kunchenguid/backpass
You don't write AGENTS.md. You train it with gradient descent.
Older snapshot: check the original source for current information.
From the publisher
README & documentation
Source previewRead the project’s overview, installation instructions and usage examples. The original README is the source of truth.
backpass Gradient descent for your agent memory. This blog post explains the why and how. backpass helps you improve your AGENTS.md, CLAUDE.md and skills with scientific rigor. It finds the agent sessions that actually ran in your repo, reads what happened in them, and proposes evidence-backed edits to your memory surface - the memory file and project skills - under a token budget, gated by you. locally or over SSH to your own machines. No API, no upload; transcripts never leave your machines except into an agent you already authenticated, and obvious secrets are redacted before they do. and every add, rewrite, or remove edit needs evidence from at least two distinct sessions. Small, noisy, bounded steps - not a rewrite. See Apply - the human gate for review and explicit scripted decisions. One run is one bounded gradient step. Requires Node >= 22.5 and acpx on your PATH. The…
Read the full README ↗ · Preview checked 2026-10-29T14:34:34.114Z
Inside the original README — Document outline
- Quick Start
- User-level memory
- Your other machines
- One file instead of the whole surface
- How It Works
- 1. Collect samples - which sessions belong to this repo
- 2. Distillation - cheap first
- 3. Calculate loss - one cheap call per transcript
- 4. Aggregate gradients - and one judged consolidation call
- 5. Gradient descent - native edits
- 6. The budget - "model size"
- 7. Skills as overflow
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Documentation belongs to its respective authors. Reported project/model license: MIT. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
What this repository does
You don't write AGENTS.md. You train it with gradient descent.
Repository facts
- Owner
- kunchenguid
- Primary language
- JavaScript
- Stars
- 1,057
- Forks
- 79
- Open issues + pull requests
- 22
- License
- MIT
- Archived
- No
- Default branch
- main
- Created
- 2026-08-21T21:12:31.000Z
- Last push
- 2026-09-19T13:51:00.000Z
Topics and intended use
No topics were included in the latest source metadata.
Review the README for scope, installation, examples and limitations. We do not run repository code or certify it.
Evaluate before installing
Review licensing and dependencies, inspect recent commits and unresolved issues, and test in an isolated environment before production use. Stars and forks alone cannot answer these questions.
Source and freshness
Source: GitHub. Metadata observed 2026-09-20T06:40:47.227Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.