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 go.honcho.dev/qANJtrD 🫡

Spin up a full Honcho with seeded data in about 15 seconds. No API key, no model spend. We made this tool for internal testing, but any one can access. The sandbox runs the whole stack against a deterministic mock model, with a workspace, peers, a session, messages and conclusions already in it. Wreck it, then reset to the clean seed in under a second, with no re-derivation and no LLM calls. Use it for a first look, for CI, for a harness smoke test, or for a demo you can rebuild between runs. Two commands, in the repo at go.honcho.dev/3jWtGeK Want real reasoning on the same seed? sandbox.sh up --provider real, with your key. Same seven seeded facts, and you can watch the deductive conclusions form on top of them.

Honcho 3.4 is out. You can now give an agent an API key that reaches one named scope and nothing else. It reads that scope's memory, adds new sessions to it, and can't see the rest of your workspace. Hand one to a client's agent, a contractor's bot, or a single project, and it shares exactly that slice of memory. Also in 3.4: 👉 Every workspace, peer, and session carries a uid that never changes, so the references you store stay put if names become editable 👉 Self-hosters can turn on working memory for dreams: the dreamer sees its remaining budget, a map of what memory holds, and notes it carries from one dream to the next 👉 A .chat() answer that misses your response_format schema returns a 422 that says so Upgrading: 3.4 runs a migration that builds five indexes in the background. On a large or busy database, build them ahead of time with the statements in the troubleshooting guide. Release notes: go.honcho.dev/SPkIrSS 🫡

Connect @bot to your other harnesses in less than 2 mins 👀

Self-hosting Honcho for a team? lovethatbrandx built Hombre, a web UI with admin/editor/viewer roles and audit logging, so more than one person can work in the same instance. Browse workspaces, peers, sessions and conclusions, chat against a peer at adjustable reasoning depth, and stream your Honcho server's logs live instead of SSH-ing into a container. Docker or from source, AGPL-3.0: github.com/lovethatbrandx/ho… Three outside contributors have already shipped PRs. Built a Honcho tool of your own? Post it in #showcase on the Honcho Discord 🫡

Running a crew of Grok Bots? Ask Honcho what each one has been doing. Every Bot is its own peer in Honcho, with its own sessions for conversations, group chats and routine runs. Ask "what did my research Bot work on this week?" and the answer comes from that Bot's actual history, while every Bot still shares one picture of you. Set it up by pasting this to any Bot: hey can you use this plugin for setting up grok bot: honcho.dev/docs/v3/guides/in… 🫡

Honcho, self hosted alongside my Hermes agent. I chose Honcho because someone else said it's amazing and after period of time where it 'got to know me' my workflows clearly improved 'under the hood'.

Give pi a memory that carries across every session and every project. The official Honcho plugin for pi loads your peer card and a summary of the session when you start, then asks Honcho about each message before pi answers, so what it knows about you shows up on the turn where you need it. pi install npm:@honcho-ai/pi-honcho Then run /honcho login inside pi. Sign in from your browser, with a device code over SSH, or with an API key. How it works: 👉 A collapsible ◆ honcho entry shows exactly what memory was loaded, every turn 👉 One session per directory by default, or one per git branch or per pi session 👉 pi gets honcho_chat and honcho_search tools, and can ask Honcho up to 5 questions at once 👉 /honcho off pauses it, and nothing is loaded or saved until you turn it back on Source and setup: go.honcho.dev/fAqtxfF 🫡

Give every Grok Bot on your account a shared memory of you. Paste this to any @bot: hey can you use this plugin for setting up grok bot: github.com/plastic-labs/honc… The Bot connects to Honcho over MCP. From then on: 👉 It reads what Honcho knows about you before it starts a task 👉 Every conversation is saved to Honcho, so the next task on a fresh machine picks up where the last one left off 👉 When you state a preference, it's saved right away 👉 You're one person to every Bot, so what one learns, the rest know too Guide: go.honcho.dev/srmADuh 🫡

Honcho 3.3 is out. You can now tell each part of Honcho's memory what matters for your app. Custom instructions reach every agent: session summaries, .chat() answers, dreams and peer cards. Set one shared instruction for a workspace or session, or give each agent its own. A coaching app could have summaries track goals and peer cards keep injuries and commitments. Also in 3.3: 👉 ChromaDB as a vector store for self-hosters, on your own Chroma server or Chroma Cloud 👉 Faster message search on large deployments, backed by a new index 👉 More headroom for session summaries of dense conversations Release notes: go.honcho.dev/L7YakaP 🫡

HELL YEAH Honcho!

Built something with Honcho? We want to see it. 👀 We're featuring builders on this account: agents, apps, plugins, automations, weird little projects that turned into something useful. Show us what it remembers, how you wired it up, and what broke along the way. You don't need a big following. Just something worth showing. Reply with what you're building or send us a DM.

Running Hermes with Honcho? Now you can see what it remembers about you, and fix it, without leaving the chat. @disapprove built Honcho Memory for Hermes, a Hermes Desktop plugin that docks a memory pane beside the conversation you're in: 👉 Memory: every conclusion from the session, newest first, with search. Pick one to see the premises behind it and what was derived from it. 👉 Ask: a question about yourself at the reasoning level you choose, scoped to this session or all of them. 👉 Context: exactly what Honcho would hand your agent right now, at a token budget you pick. 👉 Corrections: add a fact, correct a wrong conclusion, or drop a file into the session. Each one asks you to confirm, then reads the saved result back before it says done. It follows whichever chat you're focused on, across profiles, SSH and remote gateways. No telemetry, and it never touches your credentials. Open source, needs Hermes 0.21.4 or newer with Hermes Desktop. Source and setup notes at go.honcho.dev/KNQpIH8, or one install: hermes plugins install outpoints/hermes-honcho-plugin --enable

Same. I thought honcho was hooked up but the embedding model was down and my agent sucked. Turned the embeddings back on and suddenly my agent is great again

you guys should go try out Honcho if you want a good memory system for your agents. I installed Hermes on another computer recently and I forgot to setup my Honcho stuff, and I have to say, having memory off made my agent way worse to work with. honcho gang

The next Honcho happy hour is Thursday 10/22 🫡 Join us 6-8pm in SoHo with the people building & using AI agents in NYC. Drinks and pizza on us, no panel, no demos. Grab a spot at go.honcho.dev/hh and come argue about memory with the people who work on it all day.

Self-hosting Honcho? @PiotrJarecki has his Hermes agent watch our server releases and send him the migration notes on Telegram for each new version 🫡

See exactly what your coding agent remembers, and decide where it applies. With Honcho in Kilo Code, each prompt pulls in what Honcho knows that's relevant to it. You can check all of it: 👉 /honcho:recall shows the exact memory text the model received 👉 The Honcho section in Kilo's sidebar links to the session in the Honcho dashboard, with every saved message and conclusion 👉 Sessions follow your working directory by default. Switch to one per repo, per git branch, or one global session with a single line of config Works in the Kilo CLI, the desktop app, VS Code and JetBrains. Setup guide: go.honcho.dev/EaGEjiT 🫡

New to Hermes Agent? @bartslodyczka goes from a blank 16GB Mac Mini to a working agent in 20 minutes, no code. At 9:07 he adds Honcho, so his agent starts building a picture of him that carries into every chat after. Watch the full build at go.honcho.dev/qANJtrD 🫡

See what Honcho knows about you, any time. Honcho Dashboard > Explore > Chat.

Hermes keeps MEMORY.md and USER.md. Honcho sits alongside them and adds the part that needs a model: as you talk, it works out what you prefer, what you’re working on and how you like things done, then loads that back in before your next message. Session history remembers this conversation. Honcho remembers what should still matter next month. It’s one install from the @NousResearch Hermes plugin catalog: hermes plugins install honcho hermes memory setup Pick honcho from the provider list and you’re set. Full guide at go.honcho.dev/1nmD64f How it works: 👉 Two layers go into every turn: who you are, and what matters right now in this session 👉 Three knobs set cost and depth: how often context refreshes, how often Honcho reasons, and how deep each pass goes 👉 Semantic search over your past conversations, plus a session summary so it knows what you already covered today And it’s portable. Point Claude Code, Codex, OpenCode or the DeepSeek Harness at the same Honcho workspace and they can read what Hermes learned about you 🫡