llmbrain.dev

llmbrain.dev · remote MCP server · beta

The memory your coding agents were missing.

One persistent brain across every project. Claude Code and Codex read it before starting, update it before ending — and never start from zero again.

curl -sSL https://server.llmbrain.dev/install.sh | bash
projects:
all of them
orientation:
~400 tokens
deleted:
nothing, ever

Requires an account and API key — request a beta invite →

The problem

Two repos. One product. Zero shared memory.

Your frontend agent has never met your backend. Connect both repos to the same llmbrain project and they share the docs that matter — api-contract, data model, milestones, last sessions.

“read api-contract.md — implement the client exactly as the backend shipped it”

repo
shop-frontend
react · claude code
repo
shop-api
fastapi · codex
llmbrain project: shop
api-contract.mddata-model.mdmilestone v0.3issues #40–47last sessions

Goodbye, /docs

Your /docs folder, promoted to the cloud.

Stop scattering markdown across repos where it rots unread. llmbrain holds the canonical docs — versioned, freshness-tracked, and readable by every agent in every repo.

shop-api/docs/rotting since March
├─ architecture.md
├─ api-contract.md
├─ data-model.md
├─ setup-notes-OLD.md
└─ TODO-final-v2.md
llmbrain · project: shopLLM-maintained
architecture.mdfresh · 2d
api-contract.mdfresh · 4h
data-model.mdstale · 31d
product.mdfresh · 1d
status.mdfresh · 9m
read by: shop-frontend · shop-api · every session

Before / after

Same agent. Same you. Two very different mornings.

without llmbrain

  • Starts siloed in one repo, blind to the rest
  • No idea what was decided last week, or why
  • You paste the same architecture summary. Again.
  • Thousands of tokens burned re-establishing context

with llmbrain

  • orient() — every project, status, and open thread
  • Reads the canonical docs, knows what's stale
  • Picks up exactly where the last session stopped
  • ~400 tokens, then straight to work

A real session

One morning with llmbrain, start to finish.

  1. 09:02
    you

    “Continue on shop-api — take the next slice from v0.3.”

  2. 09:02
    agent → llmbrain

    Orients itself before touching code.

    orient() → focus: shop · next slice: issue #44
last session: #43 closed — pagination on /orders
  3. 09:03
    agent → llmbrain

    Pulls the slice and its references.

    issue.get(#44) → "webhook retries, exp backoff"
ref: ADR-007 · api-contract.md §webhooks
  4. 09:05
    agent

    Works the issue — code, tests, the actual job. llmbrain stays out of the way.

  5. 09:47
    agent → llmbrain

    Closes the slice and freshens the shared contract doc.

    issue.close(#44) · doc.update("api-contract.md")
§webhooks retry semantics → fresh
  6. 09:49
    memory written

    Tomorrow's session — in this repo or the frontend — starts from here.

    session.end() → next slice queued: issue #45

Stop re-explaining your projects to your agents.

llmbrain is in private beta. Leave your email and we’ll send an invite — plus the MCP endpoint to point your agents at.

nothing is ever deleted — including your spot in line