smoosh
Make anyone on your team an expert in your tech. Open Source
Start new thread
Kyran | K1 R1

smoosh - Make anyone on your team an expert in your tech. Open Source

by•
Turn git repos into RAG-AI optimised context. Pure bash, zero dependencies. We've been pairing smoosh with NotebookLM: No hallucinations. Technical or non-technical team members can get all the info they need without holding up productivity. Even our non-technical work is in repos now to have the same flow in reverse: everyone's got on-demand access to any info they need with total accuracy and source citation. Private by default. Agent native. Interactive CLI. No dependencies. Get smooshin

Add a comment

Replies

Best
Kyran | K1 R1
Maker
📌
Why smoosh? Getting codebase context into AI takes time. Other tools do this but require bloated node_modules or Python environments just to concatenate text. We built smoosh internally for a zero-dependency, native approach. It turns a 20-minute chore into one fast, reliable command. - Understand codebases: Upload to tools like NotebookLM to talk through your architecture without reading source. - Give AI context: Drop into Claude/ChatGPT for an assistant that knows your code, eliminating hallucinated APIs. - Onboard instantly: Give new hires a searchable snapshot to learn from, regardless of their technical background. - Ground your agents: Output is RAG-optimised, chunked within limits, and retains metadata. - Private by default: Runs locally. No API keys, zero telemetry. How it Works smoosh isn't just a wrapper around cat. It is a strict, structured pipeline: 1. Discovery: Uses git ls-files to perfectly respect your .gitignore. 2. Filtering: Applies extension rules (--docs, --code) or MIME-type checks (--all) to drop binaries and noise. 3. Chunking: Streams content through a fast word-count heuristic, splitting files sequentially without breaking mid-file. 4. Verification: The final output is strictly cross-referenced against the expected file list. Any mismatch yields an immediate exit 4. Features - File presets — --docs (md, txt, etc.), --code (docs + code), --all (excludes binaries via MIME checks). - Smart chunking — Stays within token limits (project_part1.md). - 100% verification — Exits 4 if output mismatches the git index. - Interactive mode — Guided setup on first run. - Remote repos — smoosh https://github.com/user/repo (clones & processes instantly). - Secrets detection — Warns on AWS keys, PATs, and PEM blocks. - Output formats — Markdown, text, or CDATA XML. - Table of contents — --toc generates a per-chunk file index. - Line numbers — --line-numbers for code reviews. - Agent-native — Designed for CI/Agents (--json, --no-interactive, deterministic exits).