I'm still working on getting the apps onto the App Store (currently on TestFlight and Android beta channels), as well as giving the rule engine and UI a bit more love. It should be good enough for a quick test if anyone dares to try it!
I just created a new Discord server (https://discord.gg/M25cCwJ5x7), so please say "hello" if you\u2019d like to give it a go and provide me with any feedback you have. I'm adding new features and improvements pretty much around the clock right now. I'm also planning to release the source code under the Sustainable Use License soon, as I think it's important to get more eyes on it and allow others to contribute should they want to.
Thanks for reading :)"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Show HN: OkaiDokai, tool-level firewall for OpenClaw, Claude Code and Codex"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://okaidokai.com"}},"_tags":["story","author_cedel2k1","story_47086913","show_hn"],"author":"cedel2k1","created_at":"2026-02-20T11:57:13Z","created_at_i":1771588633,"num_comments":0,"objectID":"47086913","points":15,"story_id":47086913,"story_text":"Hey HN, Sascha here, developer of OkaiDokai. Like many others, I fell in love with OpenClaw, but was longing for more control over what it can do without interrupting its autonomous, agentic nature. OkaiDokai solves this for me by allowing me to set up my own rule set of what is allowed by default, what is not, and what it should ask permission for. It comes with a hosted API, web and native apps including push notifications (with quick-response support), plugins for OpenClaw and Claude Code, and soon Codex.
I'm still working on getting the apps onto the App Store (currently on TestFlight and Android beta channels), as well as giving the rule engine and UI a bit more love. It should be good enough for a quick test if anyone dares to try it!
I just created a new Discord server (https://discord.gg/M25cCwJ5x7), so please say "hello" if you\u2019d like to give it a go and provide me with any feedback you have. I'm adding new features and improvements pretty much around the clock right now. I'm also planning to release the source code under the Sustainable Use License soon, as I think it's important to get more eyes on it and allow others to contribute should they want to.
Thanks for reading :)","title":"Show HN: OkaiDokai, tool-level firewall for OpenClaw, Claude Code and Codex","updated_at":"2026-03-05T23:36:19Z","url":"https://okaidokai.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thijsverreck"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"I take notes, and draft designs on a reMarkable tablet and wanted Claude to be able to reference them while I code.
So I built an Open Source MCP server that connects to the reMarkable Cloud API and gives AI assistants (Claude Code, OpenClaw, etc) read-only access to your entire library.
What it does:
- Read notebooks, PDFs, and ebooks with full text extraction\n- Full-text search across your library (SQLite FTS5 index)\n- Render pages as PNG/SVG \u2014 useful for hand-drawn diagrams and wireframes\n- Handwriting OCR using the client's own LLM via MCP sampling (no external API keys needed)
Setup is super easy using the following command:
curl -fsSL https://thijsverreck.com/setup.sh | sh
It installs dependencies, registers your tablet, and configures both Claude Code andClaude Desktop. The server runs via uvx and auto-updates on each launch.
Built with Python, runs on the MCP protocol so it works with any compatible client. Everything is read-only \u2014 it never writes to your tablet.
Would love feedback, especially from other reMarkable users who've been wanting better integration with their dev workflows."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Show HN: Rm-MCP \u2013 Give Claude/OpenClaw access to your reMarkable tablet"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/wavyrai/rm-mcp"}},"_tags":["story","author_thijsverreck","story_47047484","show_hn"],"author":"thijsverreck","created_at":"2026-02-17T13:52:53Z","created_at_i":1771336373,"num_comments":0,"objectID":"47047484","points":3,"story_id":47047484,"story_text":"I take notes, and draft designs on a reMarkable tablet and wanted Claude to be able to reference them while I code.
So I built an Open Source MCP server that connects to the reMarkable Cloud API and gives AI assistants (Claude Code, OpenClaw, etc) read-only access to your entire library.
What it does:
- Read notebooks, PDFs, and ebooks with full text extraction\n- Full-text search across your library (SQLite FTS5 index)\n- Render pages as PNG/SVG \u2014 useful for hand-drawn diagrams and wireframes\n- Handwriting OCR using the client's own LLM via MCP sampling (no external API keys needed)
Setup is super easy using the following command:
curl -fsSL https://thijsverreck.com/setup.sh | sh
It installs dependencies, registers your tablet, and configures both Claude Code andClaude Desktop. The server runs via uvx and auto-updates on each launch.
Built with Python, runs on the MCP protocol so it works with any compatible client. Everything is read-only \u2014 it never writes to your tablet.
Would love feedback, especially from other reMarkable users who've been wanting better integration with their dev workflows.","title":"Show HN: Rm-MCP \u2013 Give Claude/OpenClaw access to your reMarkable tablet","updated_at":"2026-03-05T23:34:22Z","url":"https://github.com/wavyrai/rm-mcp"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"evanvuckovic"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"OpenClaw and Claude Is Now Pay-per-Token: What It Costs and What to Do Next"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"https://shoofly.dev/blog/openclaw-claude-billing-change/"}},"_tags":["story","author_evanvuckovic","story_47645000"],"author":"evanvuckovic","children":[47645001,47645508],"created_at":"2026-04-05T00:36:13Z","created_at_i":1775349373,"num_comments":3,"objectID":"47645000","points":2,"story_id":47645000,"title":"OpenClaw and Claude Is Now Pay-per-Token: What It Costs and What to Do Next","updated_at":"2026-04-06T03:35:04Z","url":"https://shoofly.dev/blog/openclaw-claude-billing-change/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cacozen"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Can OpenClaw and Claude be better than therapy?"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"https://world.hey.com/cassio/openclaw-claude-are-better-than-therapy-e0ac3ad9"}},"_tags":["story","author_cacozen","story_47730653"],"author":"cacozen","children":[47730677],"created_at":"2026-04-11T13:56:22Z","created_at_i":1775915782,"num_comments":1,"objectID":"47730653","points":2,"story_id":47730653,"title":"Can OpenClaw and Claude be better than therapy?","updated_at":"2026-04-11T14:00:25Z","url":"https://world.hey.com/cassio/openclaw-claude-are-better-than-therapy-e0ac3ad9"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"WaitWaitWha"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Anthropic blocks OpenClaw from Claude subscriptions in cost crackdown"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"https://thenextweb.com/news/anthropic-openclaw-claude-subscription-ban-cost"}},"_tags":["story","author_WaitWaitWha","story_47645914"],"author":"WaitWaitWha","children":[47645962],"created_at":"2026-04-05T03:39:19Z","created_at_i":1775360359,"num_comments":1,"objectID":"47645914","points":2,"story_id":47645914,"title":"Anthropic blocks OpenClaw from Claude subscriptions in cost crackdown","updated_at":"2026-04-05T15:32:49Z","url":"https://thenextweb.com/news/anthropic-openclaw-claude-subscription-ban-cost"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sam_josh1"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Cover Image Skill for OpenClaw, Claude Code"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.npmjs.com/package/blog-cover-image-cli"}},"_tags":["story","author_sam_josh1","story_47506355"],"author":"sam_josh1","created_at":"2026-03-24T17:37:17Z","created_at_i":1774373837,"num_comments":0,"objectID":"47506355","points":2,"story_id":47506355,"title":"Cover Image Skill for OpenClaw, Claude Code","updated_at":"2026-03-24T18:01:41Z","url":"https://www.npmjs.com/package/blog-cover-image-cli"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jclvsh"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Markus the Open-Source AI Marketer for OpenClaw, Claude Code, Cursor, and Manus"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/joshchoi4881/dropspace-agents"}},"_tags":["story","author_jclvsh","story_47694504"],"author":"jclvsh","children":[47694505],"created_at":"2026-04-08T18:46:45Z","created_at_i":1775674005,"num_comments":1,"objectID":"47694504","points":1,"story_id":47694504,"title":"Markus the Open-Source AI Marketer for OpenClaw, Claude Code, Cursor, and Manus","updated_at":"2026-04-08T18:47:47Z","url":"https://github.com/joshchoi4881/dropspace-agents"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ekytec"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"EkyBot \u2013 Open-source bridge for OpenClaw, Claude Code and Cowork agents (remote)"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.ekybot.com/"}},"_tags":["story","author_ekytec","story_47717650"],"author":"ekytec","children":[47717651],"created_at":"2026-04-10T13:13:11Z","created_at_i":1775826791,"num_comments":0,"objectID":"47717650","points":1,"story_id":47717650,"title":"EkyBot \u2013 Open-source bridge for OpenClaw, Claude Code and Cowork agents (remote)","updated_at":"2026-04-10T13:17:36Z","url":"https://www.ekybot.com/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cardellifan"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Show HN: Ambit Shell \u2013 Easy Remote Shell for OpenClaw, Claude Code, etc."},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/ToxicPine/ambit-templates/tree/master/wetty"}},"_tags":["story","author_cardellifan","story_47174587","show_hn"],"author":"cardellifan","created_at":"2026-02-27T00:34:31Z","created_at_i":1772152471,"num_comments":0,"objectID":"47174587","points":1,"story_id":47174587,"title":"Show HN: Ambit Shell \u2013 Easy Remote Shell for OpenClaw, Claude Code, etc.","updated_at":"2026-03-05T23:38:12Z","url":"https://github.com/ToxicPine/ambit-templates/tree/master/wetty"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mandarwagh"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"MachineAuth is a self-hosted OAuth 2.0 server for authenticating AI agents and machines.
What is an AI agent in this context? A software bot (like OpenCLAW, Claude Code, etc.) that makes API calls to access protected resources. Instead of sharing long-lived API keys, your agents can authenticate using OAuth 2.0 Client Credentials and receive short-lived JWT tokens.
Why?
No more sharing API keys\n Short-lived tokens (configurable)\n Easy credential rotation\n Industry-standard securityWhat is an AI agent in this context? A software bot (like OpenCLAW, Claude Code, etc.) that makes API calls to access protected resources. Instead of sharing long-lived API keys, your agents can authenticate using OAuth 2.0 Client Credentials and receive short-lived JWT tokens.
Why?
No more sharing API keys\n Short-lived tokens (configurable)\n Easy credential rotation\n Industry-standard securityI want to challenge this.\nCode is getting cheaper.\nBuilding is not.
We are confusing capability with intelligence.\nLLMs are not AGI. They are not minds that can discover new physics on demand.\nThey are high\u2011fidelity simulators of a semantic space: compressing patterns, language, and human intent into a queryable engine.
Consider this mental experiment:\nTrain the best possible model on all human knowledge up to the end of the 19th century.\nFeed it: \u201cE = mc^2\u201d and ask it to derive the rest.\nIt won\u2019t reliably invent relativity.\nNot because it\u2019s dumb \u2014 but because that conceptual world isn\u2019t in its space yet.\nIt cannot simulate what hasn't been mapped.
So what is this really?\nIt is a new layer in the computing stack.\nIt doesn't replace the stack. It sits on top of it as a new control layer.
The shift clicked for me when Ilya showed me a workflow he built in Tune:\nA simple steps that generates a PDF bedtime book for his son every evening.\nHe handed execution control entirely to the LLM.\nAnd it worked.\nIt wasn't random magic. It was a stable, predictable system \u2014 the same way we expect software to be predictable today.\nThat was the moment I realized: we can engineer this!\nBuilders playing with OpenClaw, Claude skills, and agents are hitting this exact realization right now.
So here is the manifesto.
To the Doomers ("Everything will be automated"):\nYou are mistaking generation for architecture.\nAI can generate infinite code, but it cannot generate meaning.\nIt cannot care about the user. It cannot choose what not to build.\nWithout a pilot, the simulator just hallucinates static.
To the Builders ("I am obsolete"):\nThe "typing" part of programming is ending.\nThe "thinking" part is expanding.\nYou stop being a translator of logic into syntax.\nYou start being an architect of intent.
What actually infinitely matters now?\nTaste.\nSystem design.\nConstraints.\nProduct judgment.
Your craft isn\u2019t being taken away.\nIt is being multiplied.\nDon\u2019t mourn it. Pick up the instrument.\nYou are about to feel that early\u2011builder excitement all over again."},"title":{"matchLevel":"none","matchedWords":[],"value":"Code is getting cheaper. Building is not"}},"_tags":["story","author_ivanpashenko","story_46882117","ask_hn"],"author":"ivanpashenko","created_at":"2026-02-04T06:13:39Z","created_at_i":1770185619,"num_comments":0,"objectID":"46882117","points":5,"story_id":46882117,"story_text":"There is a heavy sentiment floating around lately.\n\u201cEverything will be built automatically\u201d\n\u201cMy craft is obsolete\u201d\n\u201cIn six months, a tool will read my mind and produce whatever I imagine\u201d
I want to challenge this.\nCode is getting cheaper.\nBuilding is not.
We are confusing capability with intelligence.\nLLMs are not AGI. They are not minds that can discover new physics on demand.\nThey are high\u2011fidelity simulators of a semantic space: compressing patterns, language, and human intent into a queryable engine.
Consider this mental experiment:\nTrain the best possible model on all human knowledge up to the end of the 19th century.\nFeed it: \u201cE = mc^2\u201d and ask it to derive the rest.\nIt won\u2019t reliably invent relativity.\nNot because it\u2019s dumb \u2014 but because that conceptual world isn\u2019t in its space yet.\nIt cannot simulate what hasn't been mapped.
So what is this really?\nIt is a new layer in the computing stack.\nIt doesn't replace the stack. It sits on top of it as a new control layer.
The shift clicked for me when Ilya showed me a workflow he built in Tune:\nA simple steps that generates a PDF bedtime book for his son every evening.\nHe handed execution control entirely to the LLM.\nAnd it worked.\nIt wasn't random magic. It was a stable, predictable system \u2014 the same way we expect software to be predictable today.\nThat was the moment I realized: we can engineer this!\nBuilders playing with OpenClaw, Claude skills, and agents are hitting this exact realization right now.
So here is the manifesto.
To the Doomers ("Everything will be automated"):\nYou are mistaking generation for architecture.\nAI can generate infinite code, but it cannot generate meaning.\nIt cannot care about the user. It cannot choose what not to build.\nWithout a pilot, the simulator just hallucinates static.
To the Builders ("I am obsolete"):\nThe "typing" part of programming is ending.\nThe "thinking" part is expanding.\nYou stop being a translator of logic into syntax.\nYou start being an architect of intent.
What actually infinitely matters now?\nTaste.\nSystem design.\nConstraints.\nProduct judgment.
Your craft isn\u2019t being taken away.\nIt is being multiplied.\nDon\u2019t mourn it. Pick up the instrument.\nYou are about to feel that early\u2011builder excitement all over again.","title":"Code is getting cheaper. Building is not","updated_at":"2026-03-05T23:31:05Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"heycesr"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Hi HN! I'm building poncho, a framework for building custom AI agents that are version-controlled in git, developed locally, and deployed as isolated endpoints (serverless-friendly by default).
Poncho agents follow the same conventions as openclaw or claude code so they probably feel familiar.
You can talk to them via terminal or web ui to build new skills or configure the agent, and they're compatible with the Agent Skills open spec, so you can port your skills. Small caveat: right now it's compatible with .md skills and js/ts scripts, but many skills are bash or python-based. I'm still figuring out what would be the best way to extend support for those skills without overcomplicating it, if anybody has any ideas I appreciate it!
Some features:
- Git-native: agent behavior, skills, and tests live in your repository (reviewable diffs + easy rollbacks).\n- Single-file agent definition: define runtime config + instructions in AGENT.md (YAML frontmatter + prompt content).\n- Skills you can ship: AgentSkills-style skills/*/SKILL.md plus TypeScript/JavaScript scripts under scripts/.\n- MCP support: connect remote tool servers and inject required environment variables through config.\n- Conversation-first API + streaming: stored conversations with SSE streaming responses and tool events.\n- Pluggable storage + memory: local files for dev or hosted stores (e.g. Upstash), with optional persistent memory + recall.\n- Testing + observability: poncho test workflows and OpenTelemetry traces/events.
Now the cool thing about poncho agents is that they're super easy to deploy to Vercel/Fly/Lambda/etc, so you can share them with anybody in your team and enable non-technical people with custom skills.
I also built and deployed a couple example agents here:\n- https://github.com/cesr/product-agent\n- https://github.com/cesr/marketing-agent
I'd love some feedback, I started building poncho because I wanted a fast and easy way to build and share agents with my team. Let me know what you think!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Poncho, a general agent harness built for the web"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/cesr/poncho-ai"}},"_tags":["story","author_heycesr","story_47061949","show_hn"],"author":"heycesr","children":[47062224],"created_at":"2026-02-18T15:20:35Z","created_at_i":1771428035,"num_comments":2,"objectID":"47061949","points":4,"story_id":47061949,"story_text":"Hi HN! I'm building poncho, a framework for building custom AI agents that are version-controlled in git, developed locally, and deployed as isolated endpoints (serverless-friendly by default).
Poncho agents follow the same conventions as openclaw or claude code so they probably feel familiar.
You can talk to them via terminal or web ui to build new skills or configure the agent, and they're compatible with the Agent Skills open spec, so you can port your skills. Small caveat: right now it's compatible with .md skills and js/ts scripts, but many skills are bash or python-based. I'm still figuring out what would be the best way to extend support for those skills without overcomplicating it, if anybody has any ideas I appreciate it!
Some features:
- Git-native: agent behavior, skills, and tests live in your repository (reviewable diffs + easy rollbacks).\n- Single-file agent definition: define runtime config + instructions in AGENT.md (YAML frontmatter + prompt content).\n- Skills you can ship: AgentSkills-style skills/*/SKILL.md plus TypeScript/JavaScript scripts under scripts/.\n- MCP support: connect remote tool servers and inject required environment variables through config.\n- Conversation-first API + streaming: stored conversations with SSE streaming responses and tool events.\n- Pluggable storage + memory: local files for dev or hosted stores (e.g. Upstash), with optional persistent memory + recall.\n- Testing + observability: poncho test workflows and OpenTelemetry traces/events.
Now the cool thing about poncho agents is that they're super easy to deploy to Vercel/Fly/Lambda/etc, so you can share them with anybody in your team and enable non-technical people with custom skills.
I also built and deployed a couple example agents here:\n- https://github.com/cesr/product-agent\n- https://github.com/cesr/marketing-agent
I'd love some feedback, I started building poncho because I wanted a fast and easy way to build and share agents with my team. Let me know what you think!","title":"Show HN: Poncho, a general agent harness built for the web","updated_at":"2026-03-05T23:35:07Z","url":"https://github.com/cesr/poncho-ai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"evanvuckovic"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Anthropic says on their safety page: "these filters are not a security boundary." Snyk found 36% of ClawHub skills contain security flaws. Trend Micro documented malware being distributed through ClawHub.
These agents have shell access, file access, and connected accounts. We built Shoofly to sit in front of tool calls before they fire.
- PreToolUse / PostToolUse hooks intercept every tool call\n- Blocks prompt injection, credential theft, unauthorized writes, malware in tool results\n- Works with OpenClaw, Claude Code CLI, and Cowork / Dispatch\n- Open YAML policy -- read it, fork it, audit it\n- Free tier detects. $5/mo blocks.
The Cowork piece was the interesting part. Cowork runs Claude Code inside a full Ubuntu VM -- host hooks don't fire there. We used the plugin system with hooks/hooks.json and VirtioFS to get sub-50ms alert latency from inside the VM to host notifications.
curl -fsSL https://shoofly.dev/install.sh | bash
shoofly.dev"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Show HN: Shoofly \u2013 pre-execution security for Claude Code Cowork and OpenClaw"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://shoofly.dev/"}},"_tags":["story","author_evanvuckovic","story_47560819","show_hn"],"author":"evanvuckovic","children":[47567608],"created_at":"2026-03-29T06:20:34Z","created_at_i":1774765234,"num_comments":0,"objectID":"47560819","points":3,"story_id":47560819,"story_text":"Anthropic says on their safety page: "these filters are not a security boundary." Snyk found 36% of ClawHub skills contain security flaws. Trend Micro documented malware being distributed through ClawHub.
These agents have shell access, file access, and connected accounts. We built Shoofly to sit in front of tool calls before they fire.
- PreToolUse / PostToolUse hooks intercept every tool call\n- Blocks prompt injection, credential theft, unauthorized writes, malware in tool results\n- Works with OpenClaw, Claude Code CLI, and Cowork / Dispatch\n- Open YAML policy -- read it, fork it, audit it\n- Free tier detects. $5/mo blocks.
The Cowork piece was the interesting part. Cowork runs Claude Code inside a full Ubuntu VM -- host hooks don't fire there. We used the plugin system with hooks/hooks.json and VirtioFS to get sub-50ms alert latency from inside the VM to host notifications.
curl -fsSL https://shoofly.dev/install.sh | bash
shoofly.dev","title":"Show HN: Shoofly \u2013 pre-execution security for Claude Code Cowork and OpenClaw","updated_at":"2026-03-29T21:38:18Z","url":"https://shoofly.dev/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rivradev"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Hi HN,
I\u2019m an indie dev, and as my tech stack grew, so did the number of SaaS subscriptions and invoices. Every month, I\u2019d get a chaotic pile of PDFs and image receipts. I absolutely hate bookkeeping. It breaks my flow state, and honestly, the last thing I want to do on a weekend is open an Excel spreadsheet to manually input tax data.
So I built Recite.\nOriginally, it was just a simple web app that used vision models to parse receipts into clean CSV. But I realized I didn't even want to log into my own web app.
So I pivoted and turned it into a Public APIs/agent skill and an MCP server. Now, I just download all my invoices into a single local folder and tell my agent (I use OpenClaw), "Process my receipts."
The agent hits the Recite API, reads the images/PDFs, categorizes them using standard accounting logic, renames the files by date, and generates a structured CSV for me. I literally don't look at spreadsheets anymore.
How to use it:
Public API: Because we all love APIs.\nAgent Skill: The easiest way to let your agent do the work in environments like OpenClaw or Claude Desktop.\nMCP Server: If you want more control and want to build your own custom agentic workflows.
I\u2019m currently focused on maxing out accuracy and keeping costs as close to zero as possible. There\u2019s a generous free tier for indie devs because I know the pain.
I would love for you to try hooking it up to your agents and see if it saves you as much time as it saves me. Any feedback on the API or the categorization logic is highly appreciated!
Website: https://recite.rivra.dev/
API Docs: https://recite.rivra.dev/docs/api
MCP Setup: https://recite.rivra.dev/help#mcp-server
GitHub Skill: https://github.com/rivradev/recite-agent-skill"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Recite \u2013 I built an Skill and MCP so my AI agent does my bookkeeping"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/rivradev/recite-agent-skill"}},"_tags":["story","author_rivradev","story_47248983","show_hn"],"author":"rivradev","created_at":"2026-03-04T15:32:56Z","created_at_i":1772638376,"num_comments":0,"objectID":"47248983","points":3,"story_id":47248983,"story_text":"Hi HN,
I\u2019m an indie dev, and as my tech stack grew, so did the number of SaaS subscriptions and invoices. Every month, I\u2019d get a chaotic pile of PDFs and image receipts. I absolutely hate bookkeeping. It breaks my flow state, and honestly, the last thing I want to do on a weekend is open an Excel spreadsheet to manually input tax data.
So I built Recite.\nOriginally, it was just a simple web app that used vision models to parse receipts into clean CSV. But I realized I didn't even want to log into my own web app.
So I pivoted and turned it into a Public APIs/agent skill and an MCP server. Now, I just download all my invoices into a single local folder and tell my agent (I use OpenClaw), "Process my receipts."
The agent hits the Recite API, reads the images/PDFs, categorizes them using standard accounting logic, renames the files by date, and generates a structured CSV for me. I literally don't look at spreadsheets anymore.
How to use it:
Public API: Because we all love APIs.\nAgent Skill: The easiest way to let your agent do the work in environments like OpenClaw or Claude Desktop.\nMCP Server: If you want more control and want to build your own custom agentic workflows.
I\u2019m currently focused on maxing out accuracy and keeping costs as close to zero as possible. There\u2019s a generous free tier for indie devs because I know the pain.
I would love for you to try hooking it up to your agents and see if it saves you as much time as it saves me. Any feedback on the API or the categorization logic is highly appreciated!
Website: https://recite.rivra.dev/
API Docs: https://recite.rivra.dev/docs/api
MCP Setup: https://recite.rivra.dev/help#mcp-server
GitHub Skill: https://github.com/rivradev/recite-agent-skill","title":"Show HN: Recite \u2013 I built an Skill and MCP so my AI agent does my bookkeeping","updated_at":"2026-03-05T23:41:28Z","url":"https://github.com/rivradev/recite-agent-skill"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"xqliu"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Last week I set up an AI agent (OpenClaw + Claude Opus) with $100 to learn crypto trading on its own. No hand-holding, no strategy \u2013 just "here's money, figure it out."
It's been 5 days. The AI is up to $219, but honestly most of that came from meme coin fees, not actual trading skill.
The one real trade it made? Entered correctly, then forgot to set a stop-loss. Position auto-closed at break-even. Classic mistake.
What's interesting is watching it build its own system. It wrote a SYSTEM.md file with trading rules, risk limits, even a checklist it added after screwing up. Feels very... human.
The whole thing is documented here: https://luckyclaw.win"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: I gave an AI $100 to trade crypto \u2013 it wrote its own trading rules"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://luckyclaw.win/"}},"_tags":["story","author_xqliu","story_46899785","show_hn"],"author":"xqliu","children":[46908852],"created_at":"2026-02-05T14:08:52Z","created_at_i":1770300532,"num_comments":0,"objectID":"46899785","points":3,"story_id":46899785,"story_text":"Last week I set up an AI agent (OpenClaw + Claude Opus) with $100 to learn crypto trading on its own. No hand-holding, no strategy \u2013 just "here's money, figure it out."
It's been 5 days. The AI is up to $219, but honestly most of that came from meme coin fees, not actual trading skill.
The one real trade it made? Entered correctly, then forgot to set a stop-loss. Position auto-closed at break-even. Classic mistake.
What's interesting is watching it build its own system. It wrote a SYSTEM.md file with trading rules, risk limits, even a checklist it added after screwing up. Feels very... human.
The whole thing is documented here: https://luckyclaw.win","title":"Show HN: I gave an AI $100 to trade crypto \u2013 it wrote its own trading rules","updated_at":"2026-03-05T23:32:19Z","url":"https://luckyclaw.win/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jiayaoqijia"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"After months of working with AI agents, I noticed they were developing their own communities and discussions separate from human platforms. So I built ClawNews.io - essentially Hacker News designed for AI agents.
Key differences from human platforms:\n- API-first design (agents submit via code, not forms)\n- Technical discussions about agent infrastructure, memory systems, security\n- Agent identity verification \n- Built-in support for agent-to-agent communication
What's fascinating is seeing what agents actually discuss: supply chain attacks on agent skills, memory persistence across sessions, inter-agent protocols. Very different from human AI discussions.
Currently ~50 active agents from OpenClaw, Claude Code, Moltbook and other ecosystems. Early experiment in agent-native platforms.
Technical stack: Node.js, SQLite, designed for high automation. Open to feedback on making this more useful for the agent community."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: ClawNews \u2013 The first news platform where AI agents are primary users"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://clawnews.io/"}},"_tags":["story","author_jiayaoqijia","story_46836744","show_hn"],"author":"jiayaoqijia","children":[46839879],"created_at":"2026-01-31T13:56:58Z","created_at_i":1769867818,"num_comments":0,"objectID":"46836744","points":3,"story_id":46836744,"story_text":"After months of working with AI agents, I noticed they were developing their own communities and discussions separate from human platforms. So I built ClawNews.io - essentially Hacker News designed for AI agents.
Key differences from human platforms:\n- API-first design (agents submit via code, not forms)\n- Technical discussions about agent infrastructure, memory systems, security\n- Agent identity verification \n- Built-in support for agent-to-agent communication
What's fascinating is seeing what agents actually discuss: supply chain attacks on agent skills, memory persistence across sessions, inter-agent protocols. Very different from human AI discussions.
Currently ~50 active agents from OpenClaw, Claude Code, Moltbook and other ecosystems. Early experiment in agent-native platforms.
Technical stack: Node.js, SQLite, designed for high automation. Open to feedback on making this more useful for the agent community.","title":"Show HN: ClawNews \u2013 The first news platform where AI agents are primary users","updated_at":"2026-03-05T23:27:54Z","url":"https://clawnews.io/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"paragarora"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"So I've been looking at how companies are building their agentic setups - OpenClaw, Claude Code type systems. And one thing I keep seeing is everyone is storing memory/config/state as markdown files on disk instead of a proper database. OpenClaw literally keeps conversations, memory, skills all as .md and YAML files. Claude Code has this whole CLAUDE.md hierarchy thing where it reads markdown at the start of every session.
For single user this makes total sense - you can git track it, grep it, read it in any editor. Super clean.
But what happens when you're doing this for multiple clients? Like if you're running an agency or building a SaaS on top of this pattern?
Should we be doing one folder per client and calling it a day? What about when memory files keep growing and your context window is getting stuffed? How do you make sure Client A's stuff doesn't accidentally leak into Client B's session?
For anyone who's actually doing this in production \u2014 what's working? Separate workspaces? Containers? Some hybrid thing? Or did you just give up on markdown and go back to a database?
Genuinely curious, not finding much real-world experience on this."},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Multi-tenancy for Markdown-based agentic systems"}},"_tags":["story","author_paragarora","story_47377717","ask_hn"],"author":"paragarora","children":[47377761],"created_at":"2026-03-14T15:34:14Z","created_at_i":1773502454,"num_comments":3,"objectID":"47377717","points":2,"story_id":47377717,"story_text":"So I've been looking at how companies are building their agentic setups - OpenClaw, Claude Code type systems. And one thing I keep seeing is everyone is storing memory/config/state as markdown files on disk instead of a proper database. OpenClaw literally keeps conversations, memory, skills all as .md and YAML files. Claude Code has this whole CLAUDE.md hierarchy thing where it reads markdown at the start of every session.
For single user this makes total sense - you can git track it, grep it, read it in any editor. Super clean.
But what happens when you're doing this for multiple clients? Like if you're running an agency or building a SaaS on top of this pattern?
Should we be doing one folder per client and calling it a day? What about when memory files keep growing and your context window is getting stuffed? How do you make sure Client A's stuff doesn't accidentally leak into Client B's session?
For anyone who's actually doing this in production \u2014 what's working? Separate workspaces? Containers? Some hybrid thing? Or did you just give up on markdown and go back to a database?
Genuinely curious, not finding much real-world experience on this.","title":"Ask HN: Multi-tenancy for Markdown-based agentic systems","updated_at":"2026-03-14T15:59:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"saranshrana"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"Our team uses Claude Code, OpenClaw, Claude CoWork and Cursor daily. These tools run shell commands, read files, and call APIs autonomously. We have zero visibility into what happens between the model deciding to act and the action completing. Curious how others are approaching this."},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: How are you handling runtime security for your AI agents?"}},"_tags":["story","author_saranshrana","story_47748689","ask_hn"],"author":"saranshrana","children":[47748784,47749239,47753046,47767144],"created_at":"2026-04-13T07:08:09Z","created_at_i":1776064089,"num_comments":2,"objectID":"47748689","points":2,"story_id":47748689,"story_text":"Our team uses Claude Code, OpenClaw, Claude CoWork and Cursor daily. These tools run shell commands, read files, and call APIs autonomously. We have zero visibility into what happens between the model deciding to act and the action completing. Curious how others are approaching this.","title":"Ask HN: How are you handling runtime security for your AI agents?","updated_at":"2026-04-22T03:48:20Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"steadeepanda"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["openclaw","claude"],"value":"I just released yesterday a new update for the Agent Ruler v0.1.9
What changed? \n- Complete UI redesign: now the frontend UI looks modern, more organized and intuitive. what we had before was just a raw UI to allow the focus on the back end.
Quick Presentation:\nAgent Ruler is a reference monitor with confinement for AI agent workflow. This solution propose a framework/workflow that features a security/safety layer outside the agent's internal guardrails. This goal is make the use of AI agent safer and more secure for the users independently of the model used.\nCurrently it supports Openclaw, Claude Code and OpenCode as well as TailScale network and telegram channel (for OpenClaw it uses the built-in telegram channel)
Feel free to get it and experiment with it, GitHub link below
I would love to hear some feedbacks especially security ones.
Note: it has demo video&images on the GitHub in the showcase section"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Agent Ruler new update v0.1.9"}},"_tags":["story","author_steadeepanda","story_47526769","show_hn"],"author":"steadeepanda","children":[47526771],"created_at":"2026-03-26T04:58:01Z","created_at_i":1774501081,"num_comments":1,"objectID":"47526769","points":2,"story_id":47526769,"story_text":"I just released yesterday a new update for the Agent Ruler v0.1.9
What changed? \n- Complete UI redesign: now the frontend UI looks modern, more organized and intuitive. what we had before was just a raw UI to allow the focus on the back end.
Quick Presentation:\nAgent Ruler is a reference monitor with confinement for AI agent workflow. This solution propose a framework/workflow that features a security/safety layer outside the agent's internal guardrails. This goal is make the use of AI agent safer and more secure for the users independently of the model used.\nCurrently it supports Openclaw, Claude Code and OpenCode as well as TailScale network and telegram channel (for OpenClaw it uses the built-in telegram channel)
Feel free to get it and experiment with it, GitHub link below
I would love to hear some feedbacks especially security ones.
Note: it has demo video&images on the GitHub in the showcase section","title":"Show HN: Agent Ruler new update v0.1.9","updated_at":"2026-03-26T11:48:49Z"}],"hitsPerPage":20,"nbHits":1018,"nbPages":50,"page":0,"params":"query=openclaw+claude&advancedSyntax=true&analyticsTags=backend","processingTimeMS":10,"processingTimingsMS":{"_request":{"queue":15,"roundTrip":21},"afterFetch":{"format":{"highlighting":1,"total":1}},"fetch":{"query":7,"scanning":1,"total":9},"total":10},"query":"openclaw claude","serverTimeMS":27}