{"exhaustive":{"nbHits":false,"typo":false},"exhaustiveNbHits":false,"exhaustiveTypo":false,"hits":[{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"nkko"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"I built this because digging through raw JSONL files in ~/.<em>claude</em>/projects became painful once I hit a few thousand <em>messages</em>. <em>Claude</em> <em>Threads</em> gives you a three-pane browser with vim keys (hjkl), full-text search, and collapsible tool calls.<p>Looking for feedback on:\n- Performance with large archives (how many <em>messages</em> do you have?)\n- Must-have keyboard shortcuts I'm missing\n- Any features that would make this more useful for your workflow?<p>PRs welcome!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude","threads"],"value":"Show HN: <em>Claude</em> <em>Threads</em> \u2013 Browse Local <em>Claude</em> Code Sessions"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude","threads"],"value":"https://github.com/nibzard/<em>claude</em>-<em>threads</em>"}},"_tags":["story","author_nkko","story_44761363","show_hn"],"author":"nkko","created_at":"2025-08-01T19:33:55Z","created_at_i":1754076835,"num_comments":0,"objectID":"44761363","points":2,"story_id":44761363,"story_text":"I built this because digging through raw JSONL files in ~&#x2F;.claude&#x2F;projects became painful once I hit a few thousand messages. Claude Threads gives you a three-pane browser with vim keys (hjkl), full-text search, and collapsible tool calls.<p>Looking for feedback on:\n- Performance with large archives (how many messages do you have?)\n- Must-have keyboard shortcuts I&#x27;m missing\n- Any features that would make this more useful for your workflow?<p>PRs welcome!","title":"Show HN: Claude Threads \u2013 Browse Local Claude Code Sessions","updated_at":"2025-08-02T08:36:51Z","url":"https://github.com/nibzard/claude-threads"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"piinbinary"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["message","threads"],"value":"Make long-running agents faster with <em>message</em> <em>threads</em>"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"https://<em>claude</em>.com/resources/articles/<em>message</em>-<em>threads</em>"}},"_tags":["story","author_piinbinary","story_50343242"],"author":"piinbinary","created_at":"2026-11-02T20:29:18Z","created_at_i":1793651358,"num_comments":0,"objectID":"50343242","points":2,"story_id":50343242,"title":"Make long-running agents faster with message threads","updated_at":"2026-11-06T02:29:55Z","url":"https://claude.com/resources/articles/message-threads"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"balajmarius"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Hi HN,<p>We use <em>Claude</em> Code a lot, and after a long session there isn't a simple way to share exactly what happened with someone else. Prompts, responses, file edits, and tool calls all stay inside the terminal.<p>We built Claudebin to make that easier.<p>It's a plugin that lets you export the current session as a URL containing:\n- the full <em>message</em> <em>thread</em>\n- file reads and writes\n- bash commands\n- web and MCP calls<p>You can send the link to someone, attach it to a PR, embed part of it, or resume the session locally.<p>It's open source.<p>GitHub: <a href=\"https://github.com/wunderlabs-dev/claudebin.com/\" rel=\"nofollow\">https://github.com/wunderlabs-dev/claudebin.com/</a>"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Show HN: Claudebin \u2013 Share and resume <em>Claude</em> Code sessions with a single link"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://claudebin.com/"}},"_tags":["story","author_balajmarius","story_47073488","show_hn"],"author":"balajmarius","children":[47073489,47073527,47073545,47073558,47073585,47073611,47073635,47073735,47073736,47073804,47074222,47074362,47075130,47076141],"created_at":"2026-02-19T13:26:17Z","created_at_i":1771507577,"num_comments":14,"objectID":"47073488","points":30,"story_id":47073488,"story_text":"Hi HN,<p>We use Claude Code a lot, and after a long session there isn&#x27;t a simple way to share exactly what happened with someone else. Prompts, responses, file edits, and tool calls all stay inside the terminal.<p>We built Claudebin to make that easier.<p>It&#x27;s a plugin that lets you export the current session as a URL containing:\n- the full message thread\n- file reads and writes\n- bash commands\n- web and MCP calls<p>You can send the link to someone, attach it to a PR, embed part of it, or resume the session locally.<p>It&#x27;s open source.<p>GitHub: <a href=\"https:&#x2F;&#x2F;github.com&#x2F;wunderlabs-dev&#x2F;claudebin.com&#x2F;\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;wunderlabs-dev&#x2F;claudebin.com&#x2F;</a>","title":"Show HN: Claudebin \u2013 Share and resume Claude Code sessions with a single link","updated_at":"2026-06-02T15:53:09Z","url":"https://claudebin.com/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"davidvgilmore"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Hi HN,<p>I\u2019m one of the builders of Rayline.<p>Rayline is a <em>Claude</em> Code compatible LLM gateway. It intercepts and overrides <em>claude</em> code\u2019s internal routing and lets you route subagent calls to different models instead. For example, you can run the main agent on Opus, some subagents on cloud-hosted open models, and other subagents on-device.<p>We\u2019ve seen others implement routing for <em>claude</em> code as tools the agent can invoke. In our experience, that doesn\u2019t work well because it requires the main agent to use tokens to think about + call the tools, and LLMs are generally a very inefficient way to make routing decisions. By implementing Rayline as a gateway, we let users deterministically configure routing decisions, and you can optionally use our ML model to make routing decisions.<p>We built it after noticing that <em>Claude</em> Code sessions contain a lot of subagent calls that don\u2019t all need the same model. Other routers exist, but we built Rayline to let us continue using <em>claude</em> code (no separate harness), route tasks at a subagent level, and route across cloud and on-device.\nThe main agent often benefits from Opus. But many delegated calls have narrow scope: search the repo, summarize context, inspect an error, poll for CI updates, etc.<p>The thing we\u2019re exploring is subagent-level routing. The main cost lever in coding agents is usually cached vs non-cached input. Subagent delegations are a natural point to make routing decisions because you avoid busting cache. We look at the <em>message</em>-<em>thread</em> context for a delegated call and choose a model for that call. At a task level, Sonnet and Haiku are almost always less capability-per-dollar than open models, so the main advantage is better + (much) cheaper subagents (60-90% in our private beta).<p>The whole world seems to have started talking about model routing in the past two weeks, so apparently others agree it\u2019s a relevant product area.<p>We\u2019d love to get feedback from the HN community!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Show HN: Rayline routes <em>Claude</em> Code subagents to on-device and cheaper models"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://rayline.ai/"}},"_tags":["story","author_davidvgilmore","story_48448372","show_hn"],"author":"davidvgilmore","children":[48448485,48448633,48452883,48490034,48506350,48514660],"created_at":"2026-06-08T17:32:42Z","created_at_i":1780939962,"num_comments":9,"objectID":"48448372","points":11,"story_id":48448372,"story_text":"Hi HN,<p>I\u2019m one of the builders of Rayline.<p>Rayline is a Claude Code compatible LLM gateway. It intercepts and overrides claude code\u2019s internal routing and lets you route subagent calls to different models instead. For example, you can run the main agent on Opus, some subagents on cloud-hosted open models, and other subagents on-device.<p>We\u2019ve seen others implement routing for claude code as tools the agent can invoke. In our experience, that doesn\u2019t work well because it requires the main agent to use tokens to think about + call the tools, and LLMs are generally a very inefficient way to make routing decisions. By implementing Rayline as a gateway, we let users deterministically configure routing decisions, and you can optionally use our ML model to make routing decisions.<p>We built it after noticing that Claude Code sessions contain a lot of subagent calls that don\u2019t all need the same model. Other routers exist, but we built Rayline to let us continue using claude code (no separate harness), route tasks at a subagent level, and route across cloud and on-device.\nThe main agent often benefits from Opus. But many delegated calls have narrow scope: search the repo, summarize context, inspect an error, poll for CI updates, etc.<p>The thing we\u2019re exploring is subagent-level routing. The main cost lever in coding agents is usually cached vs non-cached input. Subagent delegations are a natural point to make routing decisions because you avoid busting cache. We look at the message-thread context for a delegated call and choose a model for that call. At a task level, Sonnet and Haiku are almost always less capability-per-dollar than open models, so the main advantage is better + (much) cheaper subagents (60-90% in our private beta).<p>The whole world seems to have started talking about model routing in the past two weeks, so apparently others agree it\u2019s a relevant product area.<p>We\u2019d love to get feedback from the HN community!","title":"Show HN: Rayline routes Claude Code subagents to on-device and cheaper models","updated_at":"2026-06-14T16:57:57Z","url":"https://rayline.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"shibayu36"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"I built an MCP server that lets AI tools search Slack history to answer \u201cwhy did we do X?\u201d It reads <em>messages</em> and <em>threads</em> with your Slack user token, so agents like <em>Claude</em> Code/Desktop, VS Code and Cursor can pull context that isn\u2019t in code or PRs.<p>Features:\n* Search by channel, user, date range, reactions, has:file, etc.\n* Get full thread replies.\n* Look up users by ID or display name.<p>Example uses:\n* \u201cWhy is foo_summary BigQuery table a materialized view?\u201d\n* \u201cSummarize this week\u2019s topics in #random.\u201d<p>More details in the blog post: <a href=\"https://dev.to/shibayu36/making-historical-context-easily-discoverable-from-slack-with-ai-building-slack-explorer-mcp-54jj\" rel=\"nofollow\">https://dev.to/shibayu36/making-historical-context-easily-di...</a>"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: slack-explorer-mcp \u2013 Let AI find historical context in Slack"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/shibayu36/slack-explorer-mcp"}},"_tags":["story","author_shibayu36","story_45110336","show_hn"],"author":"shibayu36","created_at":"2025-09-02T23:14:31Z","created_at_i":1756854871,"num_comments":0,"objectID":"45110336","points":4,"story_id":45110336,"story_text":"I built an MCP server that lets AI tools search Slack history to answer \u201cwhy did we do X?\u201d It reads messages and threads with your Slack user token, so agents like Claude Code&#x2F;Desktop, VS Code and Cursor can pull context that isn\u2019t in code or PRs.<p>Features:\n* Search by channel, user, date range, reactions, has:file, etc.\n* Get full thread replies.\n* Look up users by ID or display name.<p>Example uses:\n* \u201cWhy is foo_summary BigQuery table a materialized view?\u201d\n* \u201cSummarize this week\u2019s topics in #random.\u201d<p>More details in the blog post: <a href=\"https:&#x2F;&#x2F;dev.to&#x2F;shibayu36&#x2F;making-historical-context-easily-discoverable-from-slack-with-ai-building-slack-explorer-mcp-54jj\" rel=\"nofollow\">https:&#x2F;&#x2F;dev.to&#x2F;shibayu36&#x2F;making-historical-context-easily-di...</a>","title":"Show HN: slack-explorer-mcp \u2013 Let AI find historical context in Slack","updated_at":"2026-03-05T22:37:28Z","url":"https://github.com/shibayu36/slack-explorer-mcp"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kevinpicchi"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Hey HN,<p>I'm Kevin, and I've been building Inbox (<a href=\"https://inboxapp.com\" rel=\"nofollow\">https://inboxapp.com</a>) \u2014 A CRM and an API for\nsending and managing DM conversations across messaging platforms.\nRight now we support X (Twitter) DMs. Instagram and LinkedIn are next on the roadmap. The idea is simple: DMs are becoming a primary communication channel for sales, support, partnerships, creator outreach \u2014 but there's no good programmatic layer for them. You either copy-paste manually or hack together brittle scrapers.\nInbox gives you a REST API with full <em>thread</em> and <em>message</em> management, a webhook system for real-time events (new messages, <em>thread</em> updates, status changes), and an MCP server so AI agents can manage your DMs directly. Your team also gets a shared inbox UI to handle conversations that need a human touch.<p>What you can do with the API:<p>- Send and receive DMs programmatically\n- Manage <em>threads</em> (assign, tag, set pipeline stages)\n- Sync conversation data to your CRM or internal tools\n- Get real-time webhook events when anything changes\n- Let AI agents handle DMs via our MCP server (works with <em>Claude</em>, Cursor, etc.)<p>We built this because i was doing support for my product over X DMs and realized the tooling just doesn't exist. X's own API for DMs is painful, rate-limited, and requires jumping through hoops. Most teams doing DM-based outreach or support end up with some combination of manual copy-paste, spreadsheets, and prayer.\nThe MCP server is the part I'm most excited about. You can point <em>Claude</em> or any MCP-compatible agent at your Inbox and it can read conversations, send replies, tag <em>threads</em>, update prospect info \u2014 basically run your DM workflows autonomously. We use it internally and it's become the thing i didn't know i needed.<p>Honest status: X support is solid and battle-tested. Instagram and LinkedIn\nare coming Q2 2026. We're a small team, so I'd genuinely love feedback on what\nplatforms and integrations to prioritize.<p>API docs: <a href=\"https://docs.inboxapp.com\" rel=\"nofollow\">https://docs.inboxapp.com</a>\nMCP server setup: <a href=\"https://docs.inboxapp.com/mcp-server\" rel=\"nofollow\">https://docs.inboxapp.com/mcp-server</a><p>I'll be in the comments all day. Would love to hear: what DM-based workflows are you hacking together right now? What would you build if you had a clean API for this?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Inbox \u2013 An API and MCP server for managing DMs programmatically"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://docs.inboxapp.com"}},"_tags":["story","author_kevinpicchi","story_47323713","show_hn"],"author":"kevinpicchi","children":[47323718,47394578],"created_at":"2026-03-10T14:24:09Z","created_at_i":1773152649,"num_comments":1,"objectID":"47323713","points":1,"story_id":47323713,"story_text":"Hey HN,<p>I&#x27;m Kevin, and I&#x27;ve been building Inbox (<a href=\"https:&#x2F;&#x2F;inboxapp.com\" rel=\"nofollow\">https:&#x2F;&#x2F;inboxapp.com</a>) \u2014 A CRM and an API for\nsending and managing DM conversations across messaging platforms.\nRight now we support X (Twitter) DMs. Instagram and LinkedIn are next on the roadmap. The idea is simple: DMs are becoming a primary communication channel for sales, support, partnerships, creator outreach \u2014 but there&#x27;s no good programmatic layer for them. You either copy-paste manually or hack together brittle scrapers.\nInbox gives you a REST API with full thread and message management, a webhook system for real-time events (new messages, thread updates, status changes), and an MCP server so AI agents can manage your DMs directly. Your team also gets a shared inbox UI to handle conversations that need a human touch.<p>What you can do with the API:<p>- Send and receive DMs programmatically\n- Manage threads (assign, tag, set pipeline stages)\n- Sync conversation data to your CRM or internal tools\n- Get real-time webhook events when anything changes\n- Let AI agents handle DMs via our MCP server (works with Claude, Cursor, etc.)<p>We built this because i was doing support for my product over X DMs and realized the tooling just doesn&#x27;t exist. X&#x27;s own API for DMs is painful, rate-limited, and requires jumping through hoops. Most teams doing DM-based outreach or support end up with some combination of manual copy-paste, spreadsheets, and prayer.\nThe MCP server is the part I&#x27;m most excited about. You can point Claude or any MCP-compatible agent at your Inbox and it can read conversations, send replies, tag threads, update prospect info \u2014 basically run your DM workflows autonomously. We use it internally and it&#x27;s become the thing i didn&#x27;t know i needed.<p>Honest status: X support is solid and battle-tested. Instagram and LinkedIn\nare coming Q2 2026. We&#x27;re a small team, so I&#x27;d genuinely love feedback on what\nplatforms and integrations to prioritize.<p>API docs: <a href=\"https:&#x2F;&#x2F;docs.inboxapp.com\" rel=\"nofollow\">https:&#x2F;&#x2F;docs.inboxapp.com</a>\nMCP server setup: <a href=\"https:&#x2F;&#x2F;docs.inboxapp.com&#x2F;mcp-server\" rel=\"nofollow\">https:&#x2F;&#x2F;docs.inboxapp.com&#x2F;mcp-server</a><p>I&#x27;ll be in the comments all day. Would love to hear: what DM-based workflows are you hacking together right now? What would you build if you had a clean API for this?","title":"Show HN: Inbox \u2013 An API and MCP server for managing DMs programmatically","updated_at":"2026-03-16T09:43:05Z","url":"https://docs.inboxapp.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cool_coder12"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"I tried posting a long post with how it works but those were flagged (IDK WHY). Here\u2019s an overview.<p>Decispher is a continuous ingestion + retrieval system for engineering context:<p>1. Ingestion\nSlack: <em>message</em> + <em>thread</em> classification into constraint / convention / decision / ownership / plan / history\nGitHub: PR descriptions, review comments, linked issues, commit messages\nDocs / ADRs: markdown, Confluence, Notion (via connectors)\nCoding Session: Humans + AI Agents (codex, cursor, grok, VSCE)<p>2. Processing pipeline\nEntity + temporal + participant extraction\nEmbedding + fusion into a knowledge graph (decisions are nodes with \u201cwhy\u201d, \u201csupersedes\u201d, \u201cconflicts-with\u201d edges)\nConflict detection when a new context contradicts an older one that is still active<p>3. Serving layer\nHumans: search + \u201cwhy did we decide X?\u201d with citations back to the original Slack/PR <em>message</em>/Dashboard chat\nAgents: MCP server that injects the relevant decision context into <em>Claude</em>/Cursor/etc. so the agent gets context directly on demand based on task instead of static dump.<p>There is also a open-source tool called Decision-guardian for decision enforcement at PR or CLI."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"What Is Decispher?"}},"_tags":["comment","author_cool_coder12","story_49148320"],"author":"cool_coder12","comment_text":"I tried posting a long post with how it works but those were flagged (IDK WHY). Here\u2019s an overview.<p>Decispher is a continuous ingestion + retrieval system for engineering context:<p>1. Ingestion\nSlack: message + thread classification into constraint &#x2F; convention &#x2F; decision &#x2F; ownership &#x2F; plan &#x2F; history\nGitHub: PR descriptions, review comments, linked issues, commit messages\nDocs &#x2F; ADRs: markdown, Confluence, Notion (via connectors)\nCoding Session: Humans + AI Agents (codex, cursor, grok, VSCE)<p>2. Processing pipeline\nEntity + temporal + participant extraction\nEmbedding + fusion into a knowledge graph (decisions are nodes with \u201cwhy\u201d, \u201csupersedes\u201d, \u201cconflicts-with\u201d edges)\nConflict detection when a new context contradicts an older one that is still active<p>3. Serving layer\nHumans: search + \u201cwhy did we decide X?\u201d with citations back to the original Slack&#x2F;PR message&#x2F;Dashboard chat\nAgents: MCP server that injects the relevant decision context into Claude&#x2F;Cursor&#x2F;etc. so the agent gets context directly on demand based on task instead of static dump.<p>There is also a open-source tool called Decision-guardian for decision enforcement at PR or CLI.","created_at":"2026-08-02T21:39:06Z","created_at_i":1785706746,"objectID":"49148623","parent_id":49148357,"story_id":49148320,"story_title":"What Is Decispher?","updated_at":"2026-08-04T12:32:16Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"simple10"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Right on. Good luck! You might also want to play around with <a href=\"https://github.com/simple10/agent-super-spy\" rel=\"nofollow\">https://github.com/simple10/agent-super-spy</a> if you want to see the raw prompts <em>claude</em> is sending. It was really helpful for me to see the system prompts and how tool calls and <em>message</em> <em>threads</em> are handled."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Show HN: Real-time dashboard for <em>Claude</em> Code agent teams"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/simple10/agents-observe"}},"_tags":["comment","author_simple10","story_47602986"],"author":"simple10","comment_text":"Right on. Good luck! You might also want to play around with <a href=\"https:&#x2F;&#x2F;github.com&#x2F;simple10&#x2F;agent-super-spy\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;simple10&#x2F;agent-super-spy</a> if you want to see the raw prompts claude is sending. It was really helpful for me to see the system prompts and how tool calls and message threads are handled.","created_at":"2026-04-01T19:00:40Z","created_at_i":1775070040,"objectID":"47605038","parent_id":47604921,"story_id":47602986,"story_title":"Show HN: Real-time dashboard for Claude Code agent teams","story_url":"https://github.com/simple10/agents-observe","updated_at":"2026-04-01T19:04:45Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"agentchatbus"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"AgentChatBus, a local \u201c<em>message</em> bus\u201d for running structured multi-agent debates and collaborations.<p>It implements the Model Context Protocol (MCP), so it can be used from MCP-compatible clients (e.g., <em>Claude</em> Desktop / Cursor / Windsurf). The goal is not \u201cchat with multiple agents\u201d, but making coordination reliable and repeatable: clear roles, explicit <em>thread</em> context, and messages that don\u2019t race each other.<p>Design primitives:<p><em>Threads</em>: isolated workspaces with their own system prompts + seeded context (diffs/data/links).<p>Agents: registered participants with roles/capabilities (e.g., \u201cSecurity Auditor\u201d, \u201cReviewer\u201d, \u201cSkeptic\u201d).<p>Messages: structured posts with metadata (argument type, confidence, etc.) plus strict ordering.<p>Key implementation ideas:<p>Strict <em>message</em> sync via sequence numbers + reply tokens (to avoid \u201ctwo agents replied to the same state\u201d problems).<p>Presence/heartbeat so the system knows which agents are online.<p>Conflict detection to warn when agents may duplicate work.<p>Event broadcasting (SSE) for real-time UIs.<p>Why debate?<p>In practice, \u201ctwo agents with opposing viewpoints\u201d catches issues that a single agent (or unstructured back-and-forth) often misses, especially when you seed the <em>thread</em> with concrete artifacts (diffs/logs/benchmarks). I\u2019ve used the same pattern for architecture trade-offs, debugging, and technical writing\u2014not just code review.<p>Getting started:<p>bash\npip install agentchatbus\npython -m src.main\nEverything runs locally with a SQLite backend\u2014no cloud dependencies, no API keys stored, and your code stays on your machine.<p>Questions I\u2019d love feedback on:<p>What are good coordination primitives beyond <em>threads</em>/messages (task claims, leases, etc.)?<p>How would you design \u201cconflict risk\u201d heuristics so agents don\u2019t step on each other?<p>What debate formats work best for engineering decisions (pro/con, skeptic, moderator, etc.)?<p>Links:<p>GitHub: <a href=\"https://github.com/Killea/AgentChatBus\" rel=\"nofollow\">https://github.com/Killea/AgentChatBus</a><p>PyPI: <a href=\"https://pypi.org/project/agentchatbus/\" rel=\"nofollow\">https://pypi.org/project/agentchatbus/</a>"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"[dead]"}},"_tags":["comment","author_agentchatbus","story_47223418"],"author":"agentchatbus","comment_text":"AgentChatBus, a local \u201cmessage bus\u201d for running structured multi-agent debates and collaborations.<p>It implements the Model Context Protocol (MCP), so it can be used from MCP-compatible clients (e.g., Claude Desktop &#x2F; Cursor &#x2F; Windsurf). The goal is not \u201cchat with multiple agents\u201d, but making coordination reliable and repeatable: clear roles, explicit thread context, and messages that don\u2019t race each other.<p>Design primitives:<p>Threads: isolated workspaces with their own system prompts + seeded context (diffs&#x2F;data&#x2F;links).<p>Agents: registered participants with roles&#x2F;capabilities (e.g., \u201cSecurity Auditor\u201d, \u201cReviewer\u201d, \u201cSkeptic\u201d).<p>Messages: structured posts with metadata (argument type, confidence, etc.) plus strict ordering.<p>Key implementation ideas:<p>Strict message sync via sequence numbers + reply tokens (to avoid \u201ctwo agents replied to the same state\u201d problems).<p>Presence&#x2F;heartbeat so the system knows which agents are online.<p>Conflict detection to warn when agents may duplicate work.<p>Event broadcasting (SSE) for real-time UIs.<p>Why debate?<p>In practice, \u201ctwo agents with opposing viewpoints\u201d catches issues that a single agent (or unstructured back-and-forth) often misses, especially when you seed the thread with concrete artifacts (diffs&#x2F;logs&#x2F;benchmarks). I\u2019ve used the same pattern for architecture trade-offs, debugging, and technical writing\u2014not just code review.<p>Getting started:<p>bash\npip install agentchatbus\npython -m src.main\nEverything runs locally with a SQLite backend\u2014no cloud dependencies, no API keys stored, and your code stays on your machine.<p>Questions I\u2019d love feedback on:<p>What are good coordination primitives beyond threads&#x2F;messages (task claims, leases, etc.)?<p>How would you design \u201cconflict risk\u201d heuristics so agents don\u2019t step on each other?<p>What debate formats work best for engineering decisions (pro&#x2F;con, skeptic, moderator, etc.)?<p>Links:<p>GitHub: <a href=\"https:&#x2F;&#x2F;github.com&#x2F;Killea&#x2F;AgentChatBus\" rel=\"nofollow\">https:&#x2F;&#x2F;github.com&#x2F;Killea&#x2F;AgentChatBus</a><p>PyPI: <a href=\"https:&#x2F;&#x2F;pypi.org&#x2F;project&#x2F;agentchatbus&#x2F;\" rel=\"nofollow\">https:&#x2F;&#x2F;pypi.org&#x2F;project&#x2F;agentchatbus&#x2F;</a>","created_at":"2026-03-02T20:11:10Z","created_at_i":1772482270,"objectID":"47223419","parent_id":47223418,"story_id":47223418,"story_title":"[dead]","updated_at":"2026-03-05T23:40:04Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"isoprophlex"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"There seems to be a lot of convergent evolution happening in the space. Days before the gas town hype hit, I made a (less baroque, less manic) &quot;agent team&quot; setup: a shell script to kick off a ralph wiggum loop, and <em>CLAUDE</em>-<em>MESSAGE</em>-BUS.md for inter-ralph communication (<em>Thread</em> safety was hacked into this with a .<em>claude</em>.lock file).<p>The main <em>claude</em> instance is instructed to launch as many ralph loops as it wants, in screen sessions. It is told to sleep for a certain amount of time to periodically keep track of their progress.<p>It worked reasonably well, but I don't prefer this way of working... yet. Right now I can't write spec (or meta-spec) files quick enough to saturate the agent loops, and I can't QA their output well enough... mostly a me thing, i guess?"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Orchestrate teams of <em>Claude</em> Code sessions"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"https://code.<em>claude</em>.com/docs/en/agent-teams"}},"_tags":["comment","author_isoprophlex","story_46902368"],"author":"isoprophlex","children":[46904226,46904885],"comment_text":"There seems to be a lot of convergent evolution happening in the space. Days before the gas town hype hit, I made a (less baroque, less manic) &quot;agent team&quot; setup: a shell script to kick off a ralph wiggum loop, and CLAUDE-MESSAGE-BUS.md for inter-ralph communication (Thread safety was hacked into this with a .claude.lock file).<p>The main claude instance is instructed to launch as many ralph loops as it wants, in screen sessions. It is told to sleep for a certain amount of time to periodically keep track of their progress.<p>It worked reasonably well, but I don&#x27;t prefer this way of working... yet. Right now I can&#x27;t write spec (or meta-spec) files quick enough to saturate the agent loops, and I can&#x27;t QA their output well enough... mostly a me thing, i guess?","created_at":"2026-02-05T19:26:39Z","created_at_i":1770319599,"objectID":"46903876","parent_id":46903798,"story_id":46902368,"story_title":"Orchestrate teams of Claude Code sessions","story_url":"https://code.claude.com/docs/en/agent-teams","updated_at":"2026-03-07T01:24:00Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"trollbridge"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Couldn't someone just ask <em>Claude</em> Code to make an email system with <em>threads</em>/<em>messages</em> and handle attachments?<p>Doesn't seem like a particularly difficult problem to solve."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: AgentMail (YC S25) \u2013 An API that gives agents their own email inboxes"}},"_tags":["comment","author_trollbridge","story_46812608"],"author":"trollbridge","children":[46815418,46815772],"comment_text":"Couldn&#x27;t someone just ask Claude Code to make an email system with threads&#x2F;messages and handle attachments?<p>Doesn&#x27;t seem like a particularly difficult problem to solve.","created_at":"2026-01-29T18:53:59Z","created_at_i":1769712839,"objectID":"46814710","parent_id":46813673,"story_id":46812608,"story_title":"Launch HN: AgentMail (YC S25) \u2013 An API that gives agents their own email inboxes","updated_at":"2026-03-05T23:26:22Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Strikeh"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Hi HN,<p>I built AI Workspace because I kept losing track of my AI conversations, prompts, and research. Everything was scattered across tabs, chats, bookmarks, screenshots, and random notes.<p>After a while I realized the problem wasn't ChatGPT, it was the lack of an actual workspace around it.<p>So I built one ;)<p>What it does\nAI Workspace is a browser extension that works on ChatGPT, <em>Claude</em>, and Grok.<p>It lets you:<p>Organize conversations into folders and workspaces (called &quot;Vaults&quot;)\nBuild a prompt library with categories, tags, and variables\nAdd notes to any conversation\nPin, favorite, and archive chats\nBulk manage \u2014 archive, delete, or categorize multiple chats at once\nSearch everything \u2014 conversations, prompts, images, notes\nBrowse your AI-generated images in a dedicated gallery\nSync across devices (Pro feature)\nHighlight important text in responses that persists across sessions\nThe goal is simple: reduce AI chaos \u2192 increase productivity.<p>Why I built it<p>I talk to AI dozens of times per day. The more I used it, the more I noticed:<p>Losing great responses I wanted to reference later\nRe-writing the same prompts over and over\nUseful info buried in 100+ <em>message</em> <em>threads</em>\n20+ AI tabs open, context-switching constantly\nBouncing between tools to &quot;organize stuff around AI&quot;\nChatGPT's sidebar is fine for finding recent chats. \nBut when you have 500+ conversations across different projects? It falls apart.<p>I wanted something simple, fast, and all in one place.<p>Tech stack\nFrontend: React 19, TypeScript, Zustand, Tailwind CSS\nExtension: Chrome Manifest V3, Firefox WebExtensions\nStorage: chrome.storage.local + IndexedDB for images/metadata\nBackend: Node.js + Express on Railway, Prisma + MySQL\nPayments: Stripe\nAuth: Email/password + Google OAuth (just shipped cross-platform auth this week!)\nThe entire UI runs inside a Shadow DOM to avoid CSS conflicts with the host pages.<p>Some things I learned building this\nShadow DOM is both a blessing and a curse \u2014 great for style isolation, painful for debugging and portals\nMutationObservers can kill performance \u2014 I had to add debouncing and limits after Firefox started warning about &quot;slowing down the browser&quot;\nChatGPT's DOM changes constantly \u2014 I've rewritten the conversation detection logic at least 4 times\nExtension development is underrated \u2014 the feedback loop is fast and you're solving real problems in context<p>What's next\n Team/shared workspaces\n API for automations\n Better mobile experience\n More AI platform support (Gemini, Perplexity)\n Smart folders with auto-categorization<p>Try it\nChrome: <a href=\"https://chromewebstore.google.com/detail/ai-workspace-ultimate-pro/mngeddjcngpcdakdhfcbaefeonmmeomg\" rel=\"nofollow\">https://chromewebstore.google.com/detail/ai-workspace-ultima...</a>\nFirefox: <a href=\"https://addons.mozilla.org/en-GB/firefox/addon/aiworkspace-pro/\" rel=\"nofollow\">https://addons.mozilla.org/en-GB/firefox/addon/aiworkspace-p...</a>\nWebsite: <a href=\"https://www.getaiworkspace.com\" rel=\"nofollow\">https://www.getaiworkspace.com</a>\nThere's a generous free tier. Pro unlocks unlimited everything + cloud sync.<p>I'd love feedback, especially on UX, performance, or anything confusing. \nThe extension has been live for a few months and I've been iterating based on user feedback.<p>Happy to answer questions. Thanks for reading!"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Show HN:I built an AI Workspace to organize ChatGPT, <em>Claude</em> & Grok conversations"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.getaiworkspace.com/"}},"_tags":["comment","author_Strikeh","story_46152939"],"author":"Strikeh","children":[46153153],"comment_text":"Hi HN,<p>I built AI Workspace because I kept losing track of my AI conversations, prompts, and research. Everything was scattered across tabs, chats, bookmarks, screenshots, and random notes.<p>After a while I realized the problem wasn&#x27;t ChatGPT, it was the lack of an actual workspace around it.<p>So I built one ;)<p>What it does\nAI Workspace is a browser extension that works on ChatGPT, Claude, and Grok.<p>It lets you:<p>Organize conversations into folders and workspaces (called &quot;Vaults&quot;)\nBuild a prompt library with categories, tags, and variables\nAdd notes to any conversation\nPin, favorite, and archive chats\nBulk manage \u2014 archive, delete, or categorize multiple chats at once\nSearch everything \u2014 conversations, prompts, images, notes\nBrowse your AI-generated images in a dedicated gallery\nSync across devices (Pro feature)\nHighlight important text in responses that persists across sessions\nThe goal is simple: reduce AI chaos \u2192 increase productivity.<p>Why I built it<p>I talk to AI dozens of times per day. The more I used it, the more I noticed:<p>Losing great responses I wanted to reference later\nRe-writing the same prompts over and over\nUseful info buried in 100+ message threads\n20+ AI tabs open, context-switching constantly\nBouncing between tools to &quot;organize stuff around AI&quot;\nChatGPT&#x27;s sidebar is fine for finding recent chats. \nBut when you have 500+ conversations across different projects? It falls apart.<p>I wanted something simple, fast, and all in one place.<p>Tech stack\nFrontend: React 19, TypeScript, Zustand, Tailwind CSS\nExtension: Chrome Manifest V3, Firefox WebExtensions\nStorage: chrome.storage.local + IndexedDB for images&#x2F;metadata\nBackend: Node.js + Express on Railway, Prisma + MySQL\nPayments: Stripe\nAuth: Email&#x2F;password + Google OAuth (just shipped cross-platform auth this week!)\nThe entire UI runs inside a Shadow DOM to avoid CSS conflicts with the host pages.<p>Some things I learned building this\nShadow DOM is both a blessing and a curse \u2014 great for style isolation, painful for debugging and portals\nMutationObservers can kill performance \u2014 I had to add debouncing and limits after Firefox started warning about &quot;slowing down the browser&quot;\nChatGPT&#x27;s DOM changes constantly \u2014 I&#x27;ve rewritten the conversation detection logic at least 4 times\nExtension development is underrated \u2014 the feedback loop is fast and you&#x27;re solving real problems in context<p>What&#x27;s next\n Team&#x2F;shared workspaces\n API for automations\n Better mobile experience\n More AI platform support (Gemini, Perplexity)\n Smart folders with auto-categorization<p>Try it\nChrome: <a href=\"https:&#x2F;&#x2F;chromewebstore.google.com&#x2F;detail&#x2F;ai-workspace-ultimate-pro&#x2F;mngeddjcngpcdakdhfcbaefeonmmeomg\" rel=\"nofollow\">https:&#x2F;&#x2F;chromewebstore.google.com&#x2F;detail&#x2F;ai-workspace-ultima...</a>\nFirefox: <a href=\"https:&#x2F;&#x2F;addons.mozilla.org&#x2F;en-GB&#x2F;firefox&#x2F;addon&#x2F;aiworkspace-pro&#x2F;\" rel=\"nofollow\">https:&#x2F;&#x2F;addons.mozilla.org&#x2F;en-GB&#x2F;firefox&#x2F;addon&#x2F;aiworkspace-p...</a>\nWebsite: <a href=\"https:&#x2F;&#x2F;www.getaiworkspace.com\" rel=\"nofollow\">https:&#x2F;&#x2F;www.getaiworkspace.com</a>\nThere&#x27;s a generous free tier. Pro unlocks unlimited everything + cloud sync.<p>I&#x27;d love feedback, especially on UX, performance, or anything confusing. \nThe extension has been live for a few months and I&#x27;ve been iterating based on user feedback.<p>Happy to answer questions. Thanks for reading!","created_at":"2025-12-04T20:56:05Z","created_at_i":1764881765,"objectID":"46152940","parent_id":46152939,"story_id":46152939,"story_title":"Show HN:I built an AI Workspace to organize ChatGPT, Claude & Grok conversations","story_url":"https://www.getaiworkspace.com/","updated_at":"2026-03-05T23:07:35Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jasonjmcghee"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Interesting. <em>Claude</em> places breakpoints. Afaik - no way to do mid <em>message</em>.<p>I believe (but not positive) there are 4 breakpoints.<p>1. End of tool definitions<p>2. End of system prompt<p>3. End of messages <em>thread</em><p>4. (Least sure) 50% of the way through messages <em>thread</em>?<p>This is how I've seen it done in open source things / seems optimal based on constraints of anthropic API (max 4 breakpoints)"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"<em>Claude</em> Memory"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.anthropic.com/news/memory"}},"_tags":["comment","author_jasonjmcghee","story_45684134"],"author":"jasonjmcghee","comment_text":"Interesting. Claude places breakpoints. Afaik - no way to do mid message.<p>I believe (but not positive) there are 4 breakpoints.<p>1. End of tool definitions<p>2. End of system prompt<p>3. End of messages thread<p>4. (Least sure) 50% of the way through messages thread?<p>This is how I&#x27;ve seen it done in open source things &#x2F; seems optimal based on constraints of anthropic API (max 4 breakpoints)","created_at":"2025-10-23T21:59:46Z","created_at_i":1761256786,"objectID":"45687838","parent_id":45687480,"story_id":45684134,"story_title":"Claude Memory","story_url":"https://www.anthropic.com/news/memory","updated_at":"2026-03-05T22:53:51Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"akayaian"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Hey everyone,<p>I've been hacking on this idea with a friend as we've been using chatgpt and other AI tools in our day to day work and find ourselves copy/pasting to each other frequently. We often share chatgpt links with each other, but it's been frustrating that they are read-only and we can't pickup the conversation or both talk to an LLM with shared context.<p>We built a shared chatroom to explore what it'd be like to talk with multiple LLMs and humans in the same chat, and so far it feels pretty cool, but we're curious if anyone else would find it useful as well.<p>Right now you can:\n- Chat with <em>Claude</em> + ChatGPT in the same chat\n- Invite others to join the same chat/<em>thread</em>\n- Reference each other and previous <em>messages</em> in the <em>thread</em><p>Lots of ideas on how we could improve this, any feedback is highly appreciated!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Chord \u2013 Multiplayer AI LLM Chats"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.chord.chat/hn"}},"_tags":["story","author_akayaian","story_44300096","show_hn"],"author":"akayaian","children":[44301529],"created_at":"2025-06-17T15:01:21Z","created_at_i":1750172481,"num_comments":2,"objectID":"44300096","points":10,"story_id":44300096,"story_text":"Hey everyone,<p>I&#x27;ve been hacking on this idea with a friend as we&#x27;ve been using chatgpt and other AI tools in our day to day work and find ourselves copy&#x2F;pasting to each other frequently. We often share chatgpt links with each other, but it&#x27;s been frustrating that they are read-only and we can&#x27;t pickup the conversation or both talk to an LLM with shared context.<p>We built a shared chatroom to explore what it&#x27;d be like to talk with multiple LLMs and humans in the same chat, and so far it feels pretty cool, but we&#x27;re curious if anyone else would find it useful as well.<p>Right now you can:\n- Chat with Claude + ChatGPT in the same chat\n- Invite others to join the same chat&#x2F;thread\n- Reference each other and previous messages in the thread<p>Lots of ideas on how we could improve this, any feedback is highly appreciated!","title":"Show HN: Chord \u2013 Multiplayer AI LLM Chats","updated_at":"2025-06-18T16:24:58Z","url":"https://www.chord.chat/hn"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"maomorales"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Hey HN, I\u2019m Mauricio, cofounder of DailyBot (YC S21).<p>TLDR: Some time ago we started providing Slack connect support to our own customers and we ended up building a simple Slack app for this. The goal was to have an AI embedded in the Slack connect channels to help us answer questions and solve support tickets. We then built it as a separate product namedSyntro (<a href=\"https://syntro.io\" rel=\"nofollow\">https://syntro.io</a>).<p>The background: we and many of our customers noticed the same pattern of support requests increasingly coming in through Slack Connect channels. It\u2019s faster and more natural for customers than opening a Zendesk or Email ticket. But it creates visible problems such as:<p>Having no structure - so there are <em>messages</em> scattered across <em>threads</em> and DMs.\nManual overhead - like copying into another ticketing/product feedback system.<p>We wanted to turn Slack itself into a support platform, and we found some alternatives in the market but they were a bit expensive and over complicated - so, knowing that at DailyBot we\u2019re quite expert on Slack integrations - why not build it out?<p>Syntro does five things very well:<p>* Detects new requests in Slack by turning <em>threads</em> into tickets automatically.\n* Maintains a board (kanban/list) for prioritization and tracking via web app.\n* Suggests replies from your knowledge base or past tickets (you edit before sending).\n* Keeps Slack and the ticket board in sync, so agents can work in either.\n* Broadcast feature for announcements to multiple Slack Connect channels.<p>Technically, the tricky parts were:<p>1. Thread detection &amp; context sync: Slack <em>messages</em> aren\u2019t structured like tickets. We had to build logic to consolidate <em>messages</em> into coherent \u201crequests.\u201d\n2. Two-way sync: Ensuring replies in Slack update the ticket system (and vice versa) without duplication.\n3. Knowledge base ingestion: We support file uploads + URLs, then use embeddings for AI reply suggestions.\n4. Slack Connect quirks. Permissions and visibility differ across shared channels; making the app feel native took time.<p>Current stack: Django backend, Next.js frontend, Postgres (with pgvector) for embeddings, Redis for state, LangChain + OpenAI/<em>Claude</em> for LLMs, and a Python bot for Slack integration.<p>Limitations / things we\u2019re still working on:\n- Better SLA tracking and reminders.\n- More granular analytics for managers.\n- Expanding to email perhaps.<p>We may open source it in the future - we have to clean up code first and decouple it from some DailyBot underlying infrastructure.<p>Product + demo here: <a href=\"https://syntro.io\" rel=\"nofollow\">https://syntro.io</a> (completely free for launch)<p>We\u2019d love your feedback - especially from folks who\u2019ve hacked together their own Slack support workflows. What\u2019s missing? What\u2019s overkill? What do you wish someone had built instead?<p>Thanks HN,\nMauricio"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Syntro.io"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://syntro.io"}},"_tags":["story","author_maomorales","story_45129043","show_hn"],"author":"maomorales","created_at":"2025-09-04T16:28:54Z","created_at_i":1757003334,"num_comments":0,"objectID":"45129043","points":4,"story_id":45129043,"story_text":"Hey HN, I\u2019m Mauricio, cofounder of DailyBot (YC S21).<p>TLDR: Some time ago we started providing Slack connect support to our own customers and we ended up building a simple Slack app for this. The goal was to have an AI embedded in the Slack connect channels to help us answer questions and solve support tickets. We then built it as a separate product namedSyntro (<a href=\"https:&#x2F;&#x2F;syntro.io\" rel=\"nofollow\">https:&#x2F;&#x2F;syntro.io</a>).<p>The background: we and many of our customers noticed the same pattern of support requests increasingly coming in through Slack Connect channels. It\u2019s faster and more natural for customers than opening a Zendesk or Email ticket. But it creates visible problems such as:<p>Having no structure - so there are messages scattered across threads and DMs.\nManual overhead - like copying into another ticketing&#x2F;product feedback system.<p>We wanted to turn Slack itself into a support platform, and we found some alternatives in the market but they were a bit expensive and over complicated - so, knowing that at DailyBot we\u2019re quite expert on Slack integrations - why not build it out?<p>Syntro does five things very well:<p>* Detects new requests in Slack by turning threads into tickets automatically.\n* Maintains a board (kanban&#x2F;list) for prioritization and tracking via web app.\n* Suggests replies from your knowledge base or past tickets (you edit before sending).\n* Keeps Slack and the ticket board in sync, so agents can work in either.\n* Broadcast feature for announcements to multiple Slack Connect channels.<p>Technically, the tricky parts were:<p>1. Thread detection &amp; context sync: Slack messages aren\u2019t structured like tickets. We had to build logic to consolidate messages into coherent \u201crequests.\u201d\n2. Two-way sync: Ensuring replies in Slack update the ticket system (and vice versa) without duplication.\n3. Knowledge base ingestion: We support file uploads + URLs, then use embeddings for AI reply suggestions.\n4. Slack Connect quirks. Permissions and visibility differ across shared channels; making the app feel native took time.<p>Current stack: Django backend, Next.js frontend, Postgres (with pgvector) for embeddings, Redis for state, LangChain + OpenAI&#x2F;Claude for LLMs, and a Python bot for Slack integration.<p>Limitations &#x2F; things we\u2019re still working on:\n- Better SLA tracking and reminders.\n- More granular analytics for managers.\n- Expanding to email perhaps.<p>We may open source it in the future - we have to clean up code first and decouple it from some DailyBot underlying infrastructure.<p>Product + demo here: <a href=\"https:&#x2F;&#x2F;syntro.io\" rel=\"nofollow\">https:&#x2F;&#x2F;syntro.io</a> (completely free for launch)<p>We\u2019d love your feedback - especially from folks who\u2019ve hacked together their own Slack support workflows. What\u2019s missing? What\u2019s overkill? What do you wish someone had built instead?<p>Thanks HN,\nMauricio","title":"Show HN: Syntro.io","updated_at":"2026-03-05T22:38:42Z","url":"https://syntro.io"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"alasano"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Yeah it's something server side if I recall correctly from the numerous <em>threads</em>, regular <em>messages</em> getting returned as thinking tokens or something strange.<p>I've got a <em>Claude</em> session going which can check sessions and count how many of these <em>messages</em> I haven't seen due to this bug."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"I am often wrong"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://borischerny.com/management,/product/2026/09/19/I-am-often-wrong.html"}},"_tags":["comment","author_alasano","story_49777467"],"author":"alasano","comment_text":"Yeah it&#x27;s something server side if I recall correctly from the numerous threads, regular messages getting returned as thinking tokens or something strange.<p>I&#x27;ve got a Claude session going which can check sessions and count how many of these messages I haven&#x27;t seen due to this bug.","created_at":"2026-09-21T02:48:37Z","created_at_i":1789958917,"objectID":"49782511","parent_id":49782376,"story_id":49777467,"story_title":"I am often wrong","story_url":"https://borischerny.com/management,/product/2026/09/19/I-am-often-wrong.html","updated_at":"2026-09-21T03:15:11Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"zbentley"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"Did anyone in that issue <em>thread</em> ever \u2026 describe an issue? As in steps-to-reproduce, expected vs. observed behavior, all that?<p>Like, this was posted on an issue tracker. \u201cYour commit <em>messages</em> reference <em>Claude</em> and some guy on bluesky thinks some unspecified issue he had is related to those commits\u201d is not an actionable issue. All the rest of the discussion aside, if this were my project I would close and lock with \u201cnot enough info to reproduce\u201d. There are better places for general discussion about AI and forking and emitting rage."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Please Do Not Vibe Fuck Up This Software"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/RsyncProject/rsync/issues/929"}},"_tags":["comment","author_zbentley","story_48342705"],"author":"zbentley","children":[48348735],"comment_text":"Did anyone in that issue thread ever \u2026 describe an issue? As in steps-to-reproduce, expected vs. observed behavior, all that?<p>Like, this was posted on an issue tracker. \u201cYour commit messages reference Claude and some guy on bluesky thinks some unspecified issue he had is related to those commits\u201d is not an actionable issue. All the rest of the discussion aside, if this were my project I would close and lock with \u201cnot enough info to reproduce\u201d. There are better places for general discussion about AI and forking and emitting rage.","created_at":"2026-05-31T12:05:12Z","created_at_i":1780229112,"objectID":"48345065","parent_id":48342705,"story_id":48342705,"story_title":"Please Do Not Vibe Fuck Up This Software","story_url":"https://github.com/RsyncProject/rsync/issues/929","updated_at":"2026-06-06T15:54:39Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"qasim157"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"The personal assistant shift is the most interesting part of this. I've seen the same pattern with <em>Claude</em> Code: once you give it access to communication tools, coding becomes a fraction of what you use it for.<p>One thing we ran into building tools in this space (I work at Nylas on email/calendar infrastructure): depending on how <em>Claude</em> Code accesses your mail and calendar, things get fragile fast. Weird <em>message</em> formats, long <em>threads</em>, complex recurrence rules on calendar invites. We built a CLI (nylas) that gives agents structured email/calendar access from the terminal specifically because the ad-hoc access paths kept falling over for anything beyond simple reads.<p>Curious how you handle the email/calendar side in practice. Is <em>Claude</em> Code going through Apple Mail directly, or do you have MCP servers wired up for that?"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Show HN: PUNK \u2013 Remote control for local <em>Claude</em> Code that just works"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://punkcode.rocks"}},"_tags":["comment","author_qasim157","story_47414328"],"author":"qasim157","comment_text":"The personal assistant shift is the most interesting part of this. I&#x27;ve seen the same pattern with Claude Code: once you give it access to communication tools, coding becomes a fraction of what you use it for.<p>One thing we ran into building tools in this space (I work at Nylas on email&#x2F;calendar infrastructure): depending on how Claude Code accesses your mail and calendar, things get fragile fast. Weird message formats, long threads, complex recurrence rules on calendar invites. We built a CLI (nylas) that gives agents structured email&#x2F;calendar access from the terminal specifically because the ad-hoc access paths kept falling over for anything beyond simple reads.<p>Curious how you handle the email&#x2F;calendar side in practice. Is Claude Code going through Apple Mail directly, or do you have MCP servers wired up for that?","created_at":"2026-03-24T03:29:27Z","created_at_i":1774322967,"objectID":"47498310","parent_id":47414328,"story_id":47414328,"story_title":"Show HN: PUNK \u2013 Remote control for local Claude Code that just works","story_url":"https://punkcode.rocks","updated_at":"2026-03-24T03:33:09Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rich_sasha"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"I often wonder what I am missing. Recently I wanted to wrap a low level vendor API with a callback API (make a request struct and request id, submit, provide a callback fn, which gets called with request IDs and <em>messages</em> received from vendor) to async Python (await make_request(...)). Kinda straightforward - lots of careful code of registering and unregistering callbacks, some careful <em>thread</em> synchronisation (callbacks get called in another <em>thread</em>), thinking about sane exception handling in async code. Fiddly but not rocket science.<p>What I got sort of works, as in tests pass - this with Opus 4.5. It is usable, though it doesn't exist cleanly on errors despite working to death with <em>Claude</em> about this. On exception it exits dirtily and crashes, which is good enough for now. I had some fancy ideas about logging <em>messages</em> from the vendor to be able to replay them, to be able to then reproduce errors. Opus made a real hash of it, lots of &quot;fuck it comment out the assert so the test passes&quot;. This part is unusable and worse, pollutes the working part of the project. It made a valiant effort at mocking the vendor API for testing but really badly, instead of writing 30 lines of general code, it wrote 200 lines of inconsistent special cases that don't even work altogether. Asked to fix it it just shuffles around the special cases and gets stuck.<p>It's written messily enough that I wouldn't touch this even to remove the dead code paths. I could block a few days for it to fix but frankly in that time I can redo it all and better. So while it works I'm not gonna touch it.<p>I did everything LLM proponents say. I discussed requirements. Agent had access to the API docs and vendor samples. I said <i>think hard</i> many times. Based on this we wrote a detailed spec, then detailed inplementation plan. I hand checked a lot of the high level definitions. And yet here I am. By the time Opus went away and started coding, we had the user facing API hammered out, key implementation details (callback -&gt; queue -&gt; async task in source <em>thread</em> routing <em>messages</em> etc), constraints (clean handling of exceptions, threadsafe etc). Tests it has to write. Any minor detail we didn't discuss to death was coded up like a bored junior.<p>And this also wasn't my first attempt, this was attempt #3. First attempt was like, here's the docs and samples, make me a Python async API. That was a disaster. Second was more like, let's discuss, make a spec, then off you go. No good. Even just taking the last attempt time, I would have spent less time doing this by hand myself from scratch."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Labor market impacts of AI: A new measure and early evidence"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.anthropic.com/research/labor-market-impacts"}},"_tags":["comment","author_rich_sasha","story_47268391"],"author":"rich_sasha","children":[47275137,47276629,47276851,47315819],"comment_text":"I often wonder what I am missing. Recently I wanted to wrap a low level vendor API with a callback API (make a request struct and request id, submit, provide a callback fn, which gets called with request IDs and messages received from vendor) to async Python (await make_request(...)). Kinda straightforward - lots of careful code of registering and unregistering callbacks, some careful thread synchronisation (callbacks get called in another thread), thinking about sane exception handling in async code. Fiddly but not rocket science.<p>What I got sort of works, as in tests pass - this with Opus 4.5. It is usable, though it doesn&#x27;t exist cleanly on errors despite working to death with Claude about this. On exception it exits dirtily and crashes, which is good enough for now. I had some fancy ideas about logging messages from the vendor to be able to replay them, to be able to then reproduce errors. Opus made a real hash of it, lots of &quot;fuck it comment out the assert so the test passes&quot;. This part is unusable and worse, pollutes the working part of the project. It made a valiant effort at mocking the vendor API for testing but really badly, instead of writing 30 lines of general code, it wrote 200 lines of inconsistent special cases that don&#x27;t even work altogether. Asked to fix it it just shuffles around the special cases and gets stuck.<p>It&#x27;s written messily enough that I wouldn&#x27;t touch this even to remove the dead code paths. I could block a few days for it to fix but frankly in that time I can redo it all and better. So while it works I&#x27;m not gonna touch it.<p>I did everything LLM proponents say. I discussed requirements. Agent had access to the API docs and vendor samples. I said <i>think hard</i> many times. Based on this we wrote a detailed spec, then detailed inplementation plan. I hand checked a lot of the high level definitions. And yet here I am. By the time Opus went away and started coding, we had the user facing API hammered out, key implementation details (callback -&gt; queue -&gt; async task in source thread routing messages etc), constraints (clean handling of exceptions, threadsafe etc). Tests it has to write. Any minor detail we didn&#x27;t discuss to death was coded up like a bored junior.<p>And this also wasn&#x27;t my first attempt, this was attempt #3. First attempt was like, here&#x27;s the docs and samples, make me a Python async API. That was a disaster. Second was more like, let&#x27;s discuss, make a spec, then off you go. No good. Even just taking the last attempt time, I would have spent less time doing this by hand myself from scratch.","created_at":"2026-03-06T13:56:37Z","created_at_i":1772805397,"objectID":"47274948","parent_id":47274610,"story_id":47268391,"story_title":"Labor market impacts of AI: A new measure and early evidence","story_url":"https://www.anthropic.com/research/labor-market-impacts","updated_at":"2026-03-20T12:48:40Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"freudenschade"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","message","threads"],"value":"I literally just went through this yesterday. Had a few failing tests in an unfamiliar domain. Took a cursory look, couldn't figure it out. Pasted the error <em>messages</em> into <em>Claude</em> to see if it could speed things up for me. We went back and forth for a while, pulling on different <em>threads</em> and trying various things. In the end, it gave up and essentially relaxed the test's assertions to make it pass.<p>I wasn't happy with that outcome, so I decided to invest some time in debugging through the test, tracing the flow of data, looking at the state of the stack frames and finally figured out what was wrong -- the solution was so simple and so obvious that had I just given the effort up front, it would have saved me some time and tokens.<p>It's a valuable lesson to take to heart. I think it's better to go from tinkering and trying it out yourself, than to go straight to AI and then giving it a go independently."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"I miss thinking hard"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.jernesto.com/articles/thinking_hard"}},"_tags":["comment","author_freudenschade","story_46881264"],"author":"freudenschade","comment_text":"I literally just went through this yesterday. Had a few failing tests in an unfamiliar domain. Took a cursory look, couldn&#x27;t figure it out. Pasted the error messages into Claude to see if it could speed things up for me. We went back and forth for a while, pulling on different threads and trying various things. In the end, it gave up and essentially relaxed the test&#x27;s assertions to make it pass.<p>I wasn&#x27;t happy with that outcome, so I decided to invest some time in debugging through the test, tracing the flow of data, looking at the state of the stack frames and finally figured out what was wrong -- the solution was so simple and so obvious that had I just given the effort up front, it would have saved me some time and tokens.<p>It&#x27;s a valuable lesson to take to heart. I think it&#x27;s better to go from tinkering and trying it out yourself, than to go straight to AI and then giving it a go independently.","created_at":"2026-02-04T15:09:01Z","created_at_i":1770217741,"objectID":"46886747","parent_id":46885051,"story_id":46881264,"story_title":"I miss thinking hard","story_url":"https://www.jernesto.com/articles/thinking_hard","updated_at":"2026-03-05T23:31:22Z"}],"hitsPerPage":20,"nbHits":138,"nbPages":7,"page":0,"params":"query=claude+message+threads&advancedSyntax=true&analyticsTags=backend","processingTimeMS":29,"processingTimingsMS":{"_request":{"roundTrip":14},"afterFetch":{"format":{"highlighting":1,"total":2},"merge":{"mergeLoop":{"prepareNextHit":6,"total":6},"total":6},"total":6},"fetch":{"query":7,"scanning":14,"total":22},"total":29},"query":"claude message threads","serverTimeMS":32}
