Rta-Smriti Brain CLI
Local project memory for AI agents: repository, sessions and decisions in SQLite with evidence labels, no cloud, no telemetry.
- 106 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Session capture is enabled by explicit consent and selected context may reach the host model
- The release is preliminary (alpha) and behaviour and schemas still change
Install
- Works in
Manual install
git clone https://github.com/sulabhdubey/rta-smriti-brain.git
cd rta-smriti-brain
python3 -m venv .venv
./.venv/bin/python -m pip install --upgrade pip
./.venv/bin/python -m pip install .The macOS and Linux block from the Ten-Minute Start section. Then `rta-brain start` for a project.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Session capture is enabled by explicit consent and selected context may reach the host model
- The release is preliminary (alpha) and behaviour and schemas still change
This is third-party code. Review the repository files before installing.
What it does
Rta-Smriti Brain answers the question every new agent session starts from scratch: what is true about this project right now. The repository, agent sessions, decisions and local evidence are stored connected, time-aware and labelled for trust (`pramana`). The `rta-brain start` command finds the Git root, creates or updates the brain, indexes the repository, starts managed sync, captures Codex sessions and opens a local console. The output is a project passport (resume, changes, decisions, handoff), bounded context packs for a new task and a graph of files, symbols, imports, calls and tests. There is verified reading of a Python definition and explainable change impact. Hosts connect through a local stdio MCP with recipes for Codex, Claude Code, Cursor, Zed, OpenCode and Gemini CLI. Capture is enabled by explicit consent and text is redacted before it is queued.
Who it is for. Developers who run long projects with several agents and do not want to retell context in every new session.
Good fit when
- A new agent session should start from a bounded context pack instead of a retelling.
- A passport with decisions and changes is assembled to hand a project to another person.
- You need to know whether a memory is confirmed and not stale.
- Capture of Codex work is enabled explicitly and stored only locally.
Not a fit when
- You need a stable version: the current release is marked alpha.
- The machine with the project has no Python 3.11 and Git.
- A rules file is enough and a separate memory base is overkill.
Example request
Continue the project work using the Rta-Smriti passport: show the latest decision, changes since the last session and a safe next step.Limitations
The current release `v1.1.0-alpha.13` is preliminary. According to the authors a three-turn Codex pilot recovered a corrected decision in a fresh session, while Claude capture and recall and cross-host handoff are not yet verified. Captured text counts as unverified history and selected context may reach your host's model provider and spend subscription or API budget. Python 3.11 or newer and Git are required; Node.js is needed only for changing the dashboard and website. The repository was last updated on 6 November 2026 and uses the MIT licence.
How to disable. Stop background services with the `rta-brain` lifecycle commands and remove the MCP entry in the host. Delete the brain folder and the virtual environment directory by hand if the data is no longer needed.
FAQ
Where is memory data stored?
Locally in SQLite on your computer. There is no cloud or telemetry and the console listens only on a loopback address.
What does `rta-brain start` do?
Finds the Git root, creates or updates the brain, indexes the repository, starts managed sync and opens a local console.
How is session capture enabled?
By explicit consent: it is scoped by source and size, text is redacted before the queue and counts as unverified until confirmed.
Which hosts are supported?
A local stdio MCP with recipes for Codex, Claude Code, Cursor, Zed, OpenCode and Gemini CLI; a Codex pilot is verified.
Are there ready binaries?
Yes, the `v1.1.0-alpha.13` release has Windows, macOS and Linux builds with an SBOM and a `SHA256SUMS.txt` file for verification.
Discussion
No questions or reviews yet.
Related
CLI toolsEncephalon stores project decisions and conventions as JSON inside Git so coding agents can search them between tasks.
Local project memory for agents: constraints, decisions and lessons live in an evidence-backed tree and return before a task.
Coding memory across sessions, machines and teammates: decisions and dead ends flow into later prompts on their own.
RoBrain passively records team and agent decisions and warns when an agent re-proposes an approach you already rejected
Local memory for agents: decisions from browser chats reach Claude Code, Cursor and Windsurf through MCP
Long-term memory for AI agents: a skill and the mindmemos command store facts from conversations and recall them in later sessions
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Rta-Smriti Brain (this card) | no | 106 | ||
| agent-memory | no | 4,315 | durable memory on Markdown files with a SQLite index as cache; here a local base with evidence labels, a project passport and a code graph | |
| agentmemory | yes | not required | 30,340 | local hook-based memory: every session is recorded and found later; here time, trust and contradictions are tracked |
README in short
The README introduces the project passport, separates verified from unverified memory and describes local privacy, architecture and release evidence in detail.