Portable memory for AI
One memory that follows you into every AI tool
Honcho is a memory layer that lives outside any single app. Connect it once, and every AI tool you use — Claude Code, Cursor, Codex, your own agent — reads the same understanding of who you are. Switch tools without starting over.
$5 free, $5 more every month with a card · Plugins for popular harnesses
What is Honcho
Your memory shouldn't live inside one app
Every AI tool keeps its own memory in its own box. ChatGPT remembers a few things. Claude reads a CLAUDE.md you wrote. Your agent has whatever you hand-wired into it. None of them talk to each other, so every time you switch tools you introduce yourself all over again.
Honcho is the layer that sits underneath all of them. You send your sessions and Honcho builds a model of who you are and how you work. Any tool you connect can then read that context and pick up where the last one left off. No copy-pasting, no re-explaining.
Honcho is genuinely really cool in how smoothly it is able to sync up context across completely different environments (Claude Code, Hermes, OpenClaw, etc). Works the same way in however many projects you decide to work on.
I am using Honcho as my memory solution for building across my laptop and PC, across Codex and Claude.
AI memory, explained
AI models don’t remember you. Memory is the part that does.
On their own, AI models are stateless. Every time you send a message, the model reads what’s in front of it and nothing else. Close the tab, start a new chat, switch tools, and it’s a blank slate again.
AI memory is the layer that changes that. It saves what happened, decides what’s worth keeping, and feeds the right pieces back in on your next turn, so the model can act like it knows you.
How it works
One command per tool, then the same memory everywhere
Honcho runs as a hosted service. Point a tool at it once and that tool can read and write your memory. Same key, same memory.
# Connect Claude Code to your Honcho memory /plugin marketplace add plastic-labs/claude-honcho /plugin install honcho@honcho
Connect your tools
Run the command in Claude Code, Cursor, or Codex. Any other client connects to the same endpoint at mcp.honcho.dev.
Just use them
Each tool now reads and updates the same memory as you work. The understanding compounds across all of them instead of resetting in each.
Official plugins for Claude Code, Cursor, and Codex, plus SDKs and a REST API for everything else. Read the docs.
Measured, not claimed
State-of-the-art performance
Set it up once. Every tool you connect reads the same you, and the memory gets sharper the more you use it, in all of them.
Benchmark figures as published at honcho.dev/evals, verifiable on GitHub. As of 2026-09-01.
Versus a file per tool
The file you keep works in exactly one place
A CLAUDE.md and a per-app memory toggle cost nothing to start. They also stay put. The moment you add a second tool, you're maintaining the same context in two places that never sync.
| A file + built-in memory | Honcho | |
|---|---|---|
| Works across tools | One tool, one machine | Every tool you connect |
| Who keeps it in sync | You, by hand, in each place | One shared memory, updated as you work |
| When it gets long | Older lines silently drop out of context | Reasons over the whole history, not the last N lines |
| Recall you never spelled out | It's not in the file, so it's gone | Infers it from how you actually work |
| Who owns it | You — it's a file | You — open source, self-hostable, your key |
Honcho doesn't replace the tools you use. It sits underneath them. Keep your CLAUDE.md if you like it; Honcho is what makes its contents travel.
Before you connect it
FAQ
What does it cost me?
Every new account gets $5 in credits, and $5 more every month once you add a payment method. The monthly $5 refreshes each billing period rather than piling up. Sending memory in is $2.00 per million tokens, reading context back is unlimited, and deeper reasoning is priced per query from $0.001 to $0.50 so you choose the depth per call. Full table at honcho.dev, as of 2026-10-01.
Is my memory actually mine?
Yes. The core is open source at github.com/plastic-labs/honcho, so you can run the whole thing yourself and keep your data on your own infrastructure. Hosted or self-hosted, it's tied to your key, not to any one app you connect.
Which tools does it work with?
Most modern AI clients. Official plugins ship for Claude Code, Cursor, and Codex; everything else connects at mcp.honcho.dev or through the Python and TypeScript SDKs and REST. Details at honcho.dev/docs.
Do I have to rip out what I already use?
No. Honcho is a layer, not a replacement. It runs alongside your existing tools, models, and files. It just gives them a shared memory to draw from. Connect one tool today; add the rest whenever.
Who is behind it?
Plastic Labs, a venture-backed team building the memory layer for AI. The benchmarks are public, the core is open source, and the founders take calls.
Start building
Bring your memory with you
Set up your context once and connect it to every tool you use. One key, one memory, everywhere you work.
Free to start · No credit card required