Install
npm install output → 940 tokens baseline → 80 tokens compressed (−91%)Install command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →Pith
Status: stable — not actively adding features. Bug fixes welcome via issues.
Token compression hooks for Claude Code. Install once, works in every session, zero config.
bash <(curl -s https://raw.githubusercontent.com/abhisekjha/pith/main/install.sh)
How it works
Four hooks attach to Claude Code's lifecycle. Every session, every project, automatically.
PostToolUse— compresses file reads, bash output, and grep results before they hit contextUserPromptSubmit— runs/pithcommands, enforces token ceiling if setSessionStart— restores compression mode, injects cache-locked rulesStop— records token usage forstatusandreport
The PostToolUse hook is where most savings happen — file reads alone account for 30–50% of context in a typical coding session.
| Tool output | What happens | Typical saving |
|---|---|---|
| File reads | skeleton only — imports, signatures, types | −88% per read |
| Bash / build | errors + summary, verbose logs discarded | −91% per run |
| Grep results | capped at 25 matches | prevents runaway searches |
| Large payloads | offloaded to ~/.pith/tmp/, pointer left in context |
prevents bloat |
Core features
1 — Auto compression (zero config)
Runs automatically. No commands needed. Every file read, bash run, and grep gets compressed on the way back to Claude.
Real session numbers:
Read large-service.ts → 1,800 tokens baseline → 210 tokens compressed (−88%)
npm install output → 940 tokens baseline → 80 tokens compressed (−91%)
2 — Output modes
Control how Claude writes back to you.
/pith → lean: drop filler, short synonyms, fragments OK
/pith ultra → maximum: abbreviate, arrows, tables
/pith precise → tight but full sentences
/pith off → disable output compression
Auto-escalation kicks in as context fills — no manual switching needed:
50% fill → LEAN
70% fill → ULTRA
85% fill → dynamic token ceiling
3 — Token status
/pith status → ASCII flow chart — baseline vs compressed vs output + plain-English savings
/pith report → interactive HTML dashboard (auto-refreshes every 30s)
4 — Smart file focus
Load only the sections of a file relevant to your current question. Saves re-reading large files.
/focus src/services/auth.ts
Returns the 5 most relevant sections with a structure overview. Works on any file type.
5 — Symbol extraction
Pull exact function or class definitions without reading the whole file.
/pith symbol src/auth.ts handleLogin → exact 30–50 lines (~95% vs full file)
/pith symbol --list src/auth.ts → all symbols with line numbers
Install
# One-liner
bash <(curl -s https://raw.githubusercontent.com/abhisekjha/pith/main/install.sh)
# From source
git clone https://github.com/abhisekjha/pith
bash pith/install.sh
Hooks install globally into ~/.claude/hooks/. Active in every session from that point on.
Update:
/pith update → git pull + re-sync hooks, commands, settings
Uninstall:
/pith uninstall
Advanced features
All present and functional. Not the daily workflow, but there when you need them.
Similar plugs
CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies
Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1
Documented system prompts from Anthropic - Claude Fable 5.1, Opus 5.5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-6-Astra, Codex. Google - Gemini 3.8 Flash, 3.1 Pro,…
Wrap Antigravity, ChatGPT Codex, Claude Code, Grok Build, Muse Code, Devin as an OpenAI/Gemini/Claude/Codex compatible API service, allowing you to enjoy the free Gemini…