Amoeba: a tool that answers a paused run's first-level questions and keeps the record
Amoeba is a tool that agents and pipelines call. When a Squadron run pauses on a question, Amoeba answers it from the project's own rules, plan and the PM's earlier answers when those settle it, and records every answer and every decline, across a program of repos (a group of related repos). Questions those sources do not settle wait for the PM. It has no process of its own, and Squadron never needs it.
For agents
If you are an agent session, use the amoeba-mcp server (docs/mcp-server.md) or the amoeba command with -j. You want it when the PM asks what is waiting, wants to answer a waiting question, wants to know what was decided while away, or wants a repo registered. Start with list (what needs the PM), then answer with a decision the PM actually gave, never one you composed. The server's own instructions carry the same map of tools.
What exists now
The amoeba command, backed by one SQLite store. Its commands (docs/cli-contract.md) are add, rm, list, handle-pause, answer, dismiss, retract, status, decisions, mark-read, report-release and changes. Each prints text for a person and, with -j, the JSON result an agent reads. amoeba add registers a repo with no other flag; amoeba list shows what needs you and what is registered. A member reads rules files with no setup (docs/rules.md). There is no resident process: each command is a short run, and Squadron never needs Amoeba.
amoeba.answering: the answering pipeline (answering.yaml), the passage cutting and citation check, the six-check decision on whether an answer stands, and the answer screen.amoeba.upstream: bounded readers of what other tools say (cf,git,sq), and the argv for starting and resuming ansqrun.amoeba.store,amoeba.operations,amoeba.cli: the store and its migrations, the operations, and the command line.amoeba.mcp:amoeba-mcp, the same operations as MCP tools for an agent session (docs/mcp-server.md: install, register, timeouts).
Where things live
- Config home,
~/.config/amoeba/:config.yaml, the one file you edit (docs/config-contract.md). - State home,
~/.local/state/amoeba/: everything Amoeba writes. The storeamoeba.sqlite3,answering/scratch, andlogs/(answering, resume,calls/<date>.jsonl). Old scratch and logs are pruned afterprune_after_days; the store's rows are kept. $AMOEBA_HOME, when set, is both homes.XDG_CONFIG_HOMEandXDG_STATE_HOMEare ignored.- An install from before slice 228 kept its state in the config home. The first command moves it, once (
docs/store-contract.md, "Moving a pre-228 install").
Installing
uv tool install amoeba-ai
The package is amoeba-ai on PyPI; it installs the amoeba and amoeba-mcp commands. Amoeba calls sq (Squadron) and cf (Context Forge), which are installed separately.
Updating the installed amoeba
From the main checkout, after a slice merges:
uv run python scripts/update_live.py update
It backs up the store and config, appends any config key the new code needs (from the config contract's example), reinstalls amoeba, migrates the store, then checks the schema version, integrity, foreign keys, member count and the amoeba-mcp tool list. If any check fails it restores the backup and the previous install itself, and exits 1. It lists the running amoeba-mcp servers at the end: they run the old code until their client sessions restart.
uv run python scripts/update_live.py restore
puts the newest backup back later. Only the latest backup set is kept, in ~/.local/state/amoeba/backups/.
The first run needs --previous <commit>, because nothing has recorded which commit the installed amoeba was built from. Every later run reads it from its own record.
Where the design lives
- Concept:
project-documents/user/project-guides/000-concept.amoeba.md - Architecture:
project-documents/user/architecture/220-arch.program-tool.md - Slice plan:
project-documents/user/architecture/220-slices.program-tool.md - The code audit that decided what stayed:
project-documents/user/analysis/220-analysis.code-audit.md
Checks
uv run ruff format --check .
uv run ruff check .
uv run pyright
uv run pytest
The live tier runs the real shipped pipeline through sq with real models, so it spends money and is never part of a bare uv run pytest. It needs sq and cf on PATH, a provider key in the shell, and these variables, which have no defaults:
AMOEBA_LIVE_ANSWER_MODEL: the Squadron model alias for the answer stepAMOEBA_LIVE_CHECK_MODEL: the alias for the check stepAMOEBA_LIVE_MAX_COST_USD: the cost cap of one run, above zero
uv run pytest tests/live/answering
Metadata
Release files for amoeba-ai 0.1.0
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 | Actions |
|---|---|---|---|---|
| amoeba_ai-0.1.0.tar.gz | 3.8 MB | With provenance | Details for amoeba_ai-0.1.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| amoeba_ai-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for amoeba_ai-0.1.0-py3-none-any.whl |
Total release size: 4.1 MB
Release files / amoeba_ai-0.1.0.tar.gz
| Download URL | amoeba_ai-0.1.0.tar.gz |
|---|---|
| Size | 3.8 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
99a650eb4fe4b886c40c13403d742c44b5b24dea6fd249514888706a41d3e32f
|
|
BLAKE2b-256 checksum How to use checksums |
f9260bb52d564f868ea318e9b639f56d2e1c46ec72e0a4f5815ca89219ec605a
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
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 Nov 15, 2026.
Transparency logRelease files / amoeba_ai-0.1.0-py3-none-any.whl
| Download URL | amoeba_ai-0.1.0-py3-none-any.whl |
|---|---|
| Size | 300.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
13b525c951340604c9688d9021efcd937e6080f902ba40ee7d0367bffd3ba25d
|
|
BLAKE2b-256 checksum How to use checksums |
da77f45fc2986a57b4d1a5d682accc002ec77c3e02fd5ee5309cc56d17bf5fac
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
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 Nov 15, 2026.
Transparency logSecurity for amoeba-ai 0.1.0
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
Both files in this release have been uploaded with provenance metadata.
Attestation for amoeba_ai-0.1.0.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 Nov 15, 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.
To learn more about protecting yourself from malicious actors, visit our package security documentation.