LOCAL-FIRST AGENTIC CODING
Small models. Scoped edits. Fewer repo fires.
Band-Aider is a local-first agentic coding harness that keeps coding models inside bounded source, test, and stabilization workflows.
pip install --pre band-aider
See the run, not just the answer.
Watch the active task, work tree, stabilization phase, queue depth, token budget, logs, and model escalation while the run is happening.
How it works
- 01 β Scope
Find the relevant files and keep the editable set narrow.
- 02 β Separate
Alpha changes source. Beta validates and repairs tests.
- 03 β Stabilize
Run targeted tests, retain failure evidence, retry, resume, and expand only when needed.
Start local. Escalate deliberately.
Use small or local OpenAI-compatible models for ordinary work. Escalation tiers are available when the run exceeds the configured modelβs limits.
The operator is still allowed to object.
Interrupt a run, inspect the exact request and response, add corrective guidance, retry without guidance, or abort.
Completion is reported, not merely declared.
Finished runs retain summaries, reports, test results, model usage, modified-file counts, and execution statistics for review.
Core capabilities
Licensing
AGPL source
Use, inspect, modify, and distribute Band-Aider under the GNU Affero General Public License v3.0.
See the AGPL repositoryCommercial license
Alternative commercial terms are available for official Band-Aider releases below version 2.0.0. Commercial licensing begins at $20.
View commercial licensesThe commercial license grants licensing rights only. Software support, maintenance, updates, and future development are not included.