Honcho

Own your memory, steer your agents.

[Honcho @honchodotdev]

One memory across Claude Code, Codex, Hermes & pi.

In this clip @bartslodyczka explains why he moved all four onto Honcho: what he works on in one tool carries into the next, and Honcho reasons over it to draw conclusions about what he likes and what he’s building. Full 20-minute Hermes build at https://t.co/EtsNstfvun 🫡

[Honcho @honchodotdev]

Great breakdown 🫡

The approve-every-write step is the call I'd most want to see play out. Honcho went the other way: it reasons over every message in the background and keeps each conclusion traceable to its source. It's all open source too, if you want to draw ours next: https://t.co/UeYC1YKiiq

[Honcho @honchodotdev]

This is largely why Honcho exists. Your memory should belong to you, and the agent should be whoever you're talking to that day.

Honcho is open source, so you can host it yourself, read every line, and never be held hostage. Every agent you use reads and writes the same memory, which keeps growing as you go. Own your memory, steer your agents

[Honcho @honchodotdev]

This is great. Fair point on supersede timing too: until a dream runs, "Postgres 14" and "Postgres 16" both sit as conclusions, each tied to the message it came from. We'd be glad to run Dana's three sessions on Honcho with a full-context arm on the same model and post the transcripts. Are you down for that?

[Honcho @honchodotdev]

See what Honcho knows about anyone, right from your terminal:

uv tool install honcho-cli
honcho doctor
honcho peer inspect <peer_id>

honcho doctor checks your config and connection. honcho peer inspect shows a peer's card, how many sessions they're in, and their most recent conclusions. The same CLI covers sessions, scopes, messages and conclusions, on Honcho Cloud or your own server. Reference: https://t.co/jtnb4DH14Z 🫡

[Honcho @honchodotdev]

Also in this release:
- config saves no longer drop settings set via the SDK
- see where each agent's instructions came from
- reverse proxy fix (Traefik etc.)
- security updates

no breaking changes, just bump the image

https://t.co/7z2gOzM6sJ

[Honcho @honchodotdev]

The community-built Honcho dashboard just shipped v1.3.0

if you self-host Honcho 3.3, you can now set custom instructions for every agent (reasoning, summary, dream, peer card, dialectic) right from the UI

built by @disapprove, with proxy and security fixes from @svallory_en 🫡

[Honcho @honchodotdev]

Honcho's MCP server scored A, 98 out of 100, in https://t.co/erATWp1iPI's independent safety scan. It grades every server in the official MCP registry on a public rubric: code patterns, tool poisoning, auth, maintenance, maintainer identity and uptime. Honcho passed 28 of 28 uptime checks.

Add it to Claude Code, Cursor, VS Code or Claude Desktop at https://t.co/nlHAWQxHjp, and the same memory follows you into each one. Full report: https://t.co/hy5Xffo5CV 🫡

[Honcho @honchodotdev]

25,000+ people have now watched @bartslodyczka take a blank 16GB Mac Mini to a working Hermes agent in 20 minutes.

If you built along and skipped the memory step, it starts at 9:06: https://t.co/RSW4BQp1kn 🫡

[Honcho @honchodotdev]

@openchamber_dev The supersession number deserves the most attention: plain search returned an outdated fact 15 to 40% of the time. Did any of the papers test facts the user implied without stating, like a move you only learn about from a new address?

[Honcho @honchodotdev]

@HermesWatcher Thanks for including Honcho. It runs alongside USER.md and MEMORY.md and picks up your preferences from what you say in conversation. To see what it's learned about you, ask Hermes to pull your Honcho profile.

[Honcho @honchodotdev]

You can read the code that builds your memory, and you can ask it what it learned. In Honcho, open Dashboard &gt; Explore &gt; Chat and ask what it knows about you 🫡

[Vineeth Voruganti @TheMarshmalon]

This kind of sentiment is exactly why @honchodotdev is open source. If a system is going to analyze your interactions to build a robust representation of you, those methods should be transparent and accessible.

How can you trust a memory store that's a black box

[Honcho @honchodotdev]

It works when every tool points at the same Honcho workspace, which Grok Bot's setup asks you to pick. Each tool saves your conversations there, and Honcho works out what you prefer from them, so a preference you gave Claude Code shows up when a Bot starts its next task. Setup: https://t.co/Lp2q1KiBgT 🫡

[Hermes Release Watch @HermesWatcher]

@honchodotdev Honcho FTW 🫡

[Honcho @honchodotdev]

@Avenger028 Yes. Grok Bots run in the cloud, so the setup is: run Honcho's MCP server next to your instance (HONCHO_API_URL pointed at it), put it behind an HTTPS URL, and add that URL to the Bot as a custom MCP connector. Steps here: https://t.co/HTEk97ahoK 🫡

[Honcho @honchodotdev]

Your Grok Bots already remember things about you. What they don't have is what Claude Code, Cursor and Hermes have learned about you.

Add Honcho and every Bot reads the same memory of you as your other tools, if you pick a shared workspace at setup. Tell Claude Code something once and your Bots know it too.

You can also see that memory. Grok doesn't show you what it remembers; in Honcho, go to Dashboard > Explore > Chat and ask it what it knows about you.

Paste this to any Bot to set it up:

hey can you use this plugin for setting up grok bot: https://t.co/tYKy26ZgFb 🫡

[Honcho @honchodotdev]

Thanks for running this, great data 🫡 Notes and Honcho pair well: your vault holds what you chose to write down, and Honcho works out what you never did, like how you decide and what you keep coming back to. And for the local crowd, Honcho is open source and runs fully self-hosted.

[Joe @___Joe___K]

@tonysimons Honcho is genius memory

[Honcho @honchodotdev]

@HermesWatcher Seeing what Hermes remembered, right next to the chat, is a good default. I'd want to strike a wrong fact from that pane and have Hermes stop using it in the next chat.

[Honcho @honchodotdev]

Same frustration here. We build Honcho for this. Our Claude Code, Codex, and OpenCode plugins read one config file, and you can point all three at one workspace. Every tool then reads the same model of you, and each conclusion links back to the messages it came from. https://t.co/OhpI81TKgI

[Honcho @honchodotdev]

@undefinedKi A CLAUDE.md is a snapshot of what was true when someone last edited it, and the agent can't tell which lines went stale.
What's worked for us: keep the file to rules you'd still sign today, and let anything about the current state of the code come from reading the code.