Stop re-explaining your codebase every session.
Give Claude Code, Cursor, and Codex a memory that survives the session. Your architecture decisions, conventions, and the gotchas you learn the hard way, in plain Markdown you own. We use this every day to build Basic Memory itself.
## The problem
Your AI coding tool forgets everything between sessions. You re-explain the same architecture, the same conventions, the same hard-won gotchas. Over and over, in every new chat and every new tool. And the moment a PR merges or a deploy goes out, the reasoning behind it starts to fade from human memory too.
## How it works in Basic Memory
## What you get
Your agent stops re-asking. Architecture decisions, API designs, deploy histories, and the reasoning behind every PR carry across every session, tool, and teammate. Readable by you, queryable by your AI, and durable past any one chat window.
## In practice
We dogfood every part of it. The CI agent below runs on the basic-memory and basic-memory-cloud repos. On every PR merge and every successful production deploy. It captures the story of what changed, in our voice, into a Basic Memory project that the whole team and our AI agents read from.
.github/workflows/basic-memory.ymlview source Runs on PR close and Deploy Production completion. Collects GitHub context, hands it to Codex with our voice guide and an output schema, and publishes the result to Basic Memory Cloud.
on:
pull_request:
types: [closed]
workflow_run:
workflows: ["Deploy Production"]
types: [completed]
# ... checkout, install bm, collect context ...
- name: Synthesize project update with Codex
uses: openai/codex-action@v1
with:
prompt-file: .github/basic-memory/memory-ci-capture.md
output-file: ${{ runner.temp }}/agent-synthesis.json
output-schema-file: ${{ runner.temp }}/agent-synthesis.schema.json
- name: Publish project update
run: bm ci publish --cloud
--config .github/basic-memory/config.yml
--context .github/basic-memory/project-update-context.json
--synthesis ${{ runner.temp }}/agent-synthesis.json.github/basic-memory/SOUL.mdview source The repo-local voice guide. Tone, taste, what to say, what to avoid. Separated from the schema so the same agent can match any team's voice.
# Auto BM Soul
Write project updates for humans who will return later
trying to understand what happened.
## Voice
- Clear, direct, warm, and technically honest.
- Prefer concrete observations over generic praise.
- It is okay to say when code is messy, risky, clever,
boring, or satisfying.
## Do Not
- Do not invent intent, impact, tests, or drama.
- Turn the note into marketing copy.
- Hide uncertainty behind confident prose.What the CI agent's output looks like, opened in Basic MemorySame PR as the quote below, rendered in the Basic Memory web app. The agent didn't fill in fields. It wrote a story (Summary, Story, Problem Addressed, How The Change Solves It, Impact On The System) in our voice, then attached the structured observations and source links underneath. Every PR merge and every prod deploy adds one of these to the team's shared knowledge base.
“This is the kind of small, durable correctness fix that prevents memory URLs from becoming false promises. Future agents should avoid rebuilding synthetic observation permalinks in service code; the model property carries the PostgreSQL length constraint and digest collision handling for a reason.”
From the Project Memory section. Written by the CI agent, in our voice, indexed in Basic Memory for every future human and AI to read before touching that code. This one is from a real basic-memory PR that closed issue #929.
see the PR## FAQ
Free and open source to run locally. Two minutes to connect your first AI tool.
Knowledge tools built to last
© 2026 Basic Memory - Basic Machines