ShellBrain
Long-term Memory for AI Agents.
ShellBrain carries useful lessons from one agent task to the next. It preserves decision reasons, failed approaches, product direction, and preferences that the resulting code does not explain.
Install
curl -L shellbrain.ai/install | bash
Works for Codex, Claude Code, and Cursor. The installer configures the runtime automatically. Repos register themselves on first use.
Requirements.
- macOS or Linux, Python 3.11+, Docker for the managed local Postgres+pgvector runtime.
Upgrade for latest capabilities
shellbrain upgrade
You can also run curl -L shellbrain.ai/upgrade | bash.
Recall in one command
Architecture
ShellBrain stores evidence and two forms of reusable knowledge:
- Episodic knowledge records evidence. It stores prompts, agent steps, tool calls, and outputs from each session.
- Empirical knowledge extracts concrete memories. It organizes problems, solutions, failed tactics, facts, preferences, and changes in a semantic graph for case-based reasoning.
- Conceptual knowledge abstracts reusable ideas. Its concept graph connects claims, relations, and implementations to empirical knowledge.
Memories and concept claims link to supporting evidence. A concept can express a product principle without a duplicate memory. Current code remains the source of truth for implementation details.
How Agents Use ShellBrain
Recall
Working agents run shellbrain recall to get one compact brief for the current task.
Recall and automatic learning use the same evidence selector. It combines BM25, vector similarity, and graph links. Learning agents receive the selected records. The recall agent can summarize them.
Recall receives only the quoted query. Include the relevant task, failure, subsystem, or decision in the question.
shellbrain recall "What is ShellBrain, and how does it help a working coding agent?"
Response format:
{
"status": "ok",
"data": {
"brief": {
"summary": "...",
"constraints": ["..."],
"known_traps": ["..."],
"prior_cases": ["..."],
"concept_orientation": ["..."],
"anchors": ["`README.md`"],
"conflicts": ["..."],
"gaps": ["..."],
"next_checks": ["..."]
},
"fallback_reason": null
},
"errors": []
}
Recall selects an evidence pack in code, then asks an inner agent to summarize it. Source attribution comes from the selected records. Provider failures return a brief directly from that same pack.
Recall provider
Codex is the default. Choose one provider for recall across repositories:
shellbrain admin recall provider inception
shellbrain admin recall provider codex
shellbrain admin recall provider claude
For faster recall with Mercury, set your Inception API key and select Inception:
export INCEPTION_API_KEY="your-inception-api-key"
shellbrain admin recall provider inception
Start your agent host with that environment. .env files are not loaded automatically. Automatic memory building is unchanged.
Selection is stored in ~/.shellbrain/recall-provider.toml (or under SHELLBRAIN_HOME). API failures return deterministic context with failure metadata. The API uses a 10-second socket timeout and no automatic retries; a slowly trickling response can exceed that elapsed time.
Memory Discipline
ShellBrain keeps memory grounded in evidence and narrow in scope. Agents request memory when they need it.
Memory that cannot justify itself should not persist.
Use ShellBrain
Use Shellbrain with your preferred agent. Then work as usual.
- Claude Code: Use
/shellbrainto recall context at task boundaries. - Codex: Use $shellbrain to recall context at task boundaries.
- Cursor: Use
/shellbrainto recall context at task boundaries.
Repair
Run shellbrain upgrade to repair an existing installation. Setup and schema migrations run automatically.
Docs
- For Humans: installation, upgrades, and first steps
- For Agents: agent workflow and memory rules
- Technical Docs: detailed documentation and code map
1 maintainer
Metadata
1 maintainer
Release files for shellbrain 0.1.60
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| shellbrain-0.1.60.tar.gz | 360.2 kB | With provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| shellbrain-0.1.60-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 881.8 kB
Release files / shellbrain-0.1.60.tar.gz
| Download URL | shellbrain-0.1.60.tar.gz |
|---|---|
| Size | 360.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
dae589afdea5fa67369ea1e0705e1c3e09fc9f3a430dbe012886b09958dea3f0
|
|
BLAKE2b-256 checksum How to use checksums |
ad664198953368f08f189572dae40370c72682e58faa397fdc3b0bab3328f930
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 23, 2026.
Transparency logRelease files / shellbrain-0.1.60-py3-none-any.whl
| Download URL | shellbrain-0.1.60-py3-none-any.whl |
|---|---|
| Size | 521.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
0968886c2629af9a31a1a79d06b8e429ffaf1c83a04f891ab7b9e7fad5c14455
|
|
BLAKE2b-256 checksum How to use checksums |
c172b91a6af449fdeb8eb309de36454cd86f3ded3144a929ccbc3bc741db3a95
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 23, 2026.
Transparency logSecurity for shellbrain 0.1.60
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
All 2 files in this release have been uploaded with provenance metadata.
Attestation for shellbrain-0.1.60.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Sep 23, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
