# teamchong/pxpipe

cut Claude Code token usage by rendering text context as images

Repository: https://github.com/teamchong/pxpipe
Canonical: https://ross.abutalabs.com/products/pxpipe
Language: TypeScript
License: MIT
License Family: permissive
Last push: 2026-08-26T17:45:53+00:00

## Health v2 (maintenance only)
Score: 80/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 98, longevity 7
- inputs: {"age_days": 105, "days_push": 7, "days_rel": 13, "gap_med": 2, "n_releases_24m": 8}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 7280, forks 634 (observed 2026-08-28T04:09:57.984404+00:00)

## What it is
pxpipe is a local TypeScript proxy that reduces Claude Code input token usage by rewriting bulky text context (system prompts, tool docs, history) into compact PNG images before requests leave the machine. It exploits the fixed token cost of images by pixel dimensions to pack ~3.1 chars per image-token versus ~1 char per text-token, cutting end-to-end bills by roughly 59-70%.

## Use cases
- reduce Claude Code API token costs
- fit more context into an LLM context window
- lower bills when using Claude Code on large codebases
- proxy Anthropic API requests through a local token optimizer
- track per-request token savings with a dashboard
- render dense code and JSON context as images for vision models

## When to choose
- you use Claude Code heavily and want to cut input token costs
- you need to fit large system prompts or tool output into a limited context window
- you want a local, MIT-licensed proxy with no data leaving your machine
- you want measurable per-request token savings via a dashboard and events log

## When to avoid
- your model or provider lacks a vision/image input channel
- your workload involves tasks where image-rendered text degrades model accuracy
- you need guaranteed exact text fidelity for critical instructions
- you use non-Anthropic APIs incompatible with the proxy's request rewriting

## Facets
- artifact type: cli-tool
- maturity: active
- function: llm-inference, proxy, developer-tools, image-processing
- domain: developer-tools, large-language-models
- platform: cli, cross-platform, windows
- tags: claude-code, token-optimization, llm-proxy, cost-reduction, vision-context, png-rendering, anthropic-api, command-line, ai-agents, nodejs, macos, linux

## Member repositories
- teamchong/pxpipe (main) score 80

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:57.984404+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:38:53.559616+00:00, confidence not recorded.
  - readme: https://github.com/teamchong/pxpipe (fetched 2026-08-28T04:09:57.984404+00:00, sha cd733db4ec9a)
- Data as of 2026-08-30T08:39:29.467469+00:00.