teamchong /
pxpipe
Local proxy that renders Claude Code's bulky context as images to cut input tokens
- 7.5k stars
- 87 stars since 3 Oct
- TypeScript
- Updated yesterday
- Some experience
- Token & cost savings
Coding agents like Claude Code re-send a lot of text on every request: the system prompt, tool documentation and old history. An image costs tokens by its pixel size, not by how much text it holds, so dense text packed into a picture can cost far less.
pxpipe runs on your machine between the agent and the API. It renders the bulky, older parts of each request into PNG pages, adds a short factsheet of precision-critical tokens, and keeps recent turns as text. Only the request changes; the model's streamed output is untouched. A dashboard shows tokens saved and each conversion side by side.
The README is open about the trade-offs. Reading images is lossy: exact hex strings are sometimes silently misread, depending on the model. It pays off on token-dense content like code and JSON and loses on sparse prose, so a profitability gate decides what to image. It reports roughly 60 to 70 percent lower bills on Claude Code traffic, checked per request against a count_tokens baseline.
What it does
Local proxy for Claude Code
Start it with npx and point ANTHROPIC_BASE_URL at it. Requests are rewritten on your machine before they leave.
Warp mode
pxpipe warp -- claudegives the same effect without setting a base URL. The README says this keeps /remote-control and claude.ai connectors working, and it also lists cursor-agent and codex.Dashboard and savings log
A local page shows tokens saved, side-by-side text and image conversions, a kill switch and model toggles. Per-request counts are written to ~/.pxpipe/events.jsonl.
Offline export
Render files, stdin or git diffs to PNG pages, a factsheet and a prompt without running the proxy. You can upload them to image-capable clients such as Cursor.
Per-model profiles and benchmarks
Different render settings for Claude, Gemini, Sol and Grok, with published scores for arithmetic, gist recall and exact-hex recall. Several models are opt-in.
What you could build
- Run Claude Code through the proxy to lower input token spend on long sessions.
- Fit more history into a context window before needing to compact.
- Export a source folder or git diff as PNG pages to paste into an image-upload client.
- Compare text and image versions of your own requests in the dashboard to check the savings.
- Wrap codex or cursor-agent with warp and a route rule to test imaging on other agents.
Get started
Start the proxy with npx, then launch Claude Code with ANTHROPIC_BASE_URL pointing at it, or use pxpipe warp instead. Open http://127.0.0.1:47821/ to watch the savings.
npx pxpipe-proxyANTHROPIC_BASE_URL=http://127.0.0.1:47821 claudepxpipe warp -- claudeCommands are copied from the project's README. Read it on GitHub for the full setup.
More in Token & cost savings
caveman
Makes coding agents answer tersely and shrinks the logs and output they read
rtk
A command-line proxy that shrinks shell output before your AI coding agent reads it
headroom
Shrinks tool outputs, logs, and files before your AI agent sends them to the model
codegraph
A local code knowledge graph that lets AI coding agents look up code in one call
freellmapi
One OpenAI-style API that routes across dozens of free LLM provider tiers
context-mode
Keeps bulky tool output out of your coding agent's context window
9router
Local proxy that routes AI coding tools to subscription, cheap, or free models
Switchyard
Routing library that picks which LLM handles each agent request
billion-context
A proxy that compresses long coding-agent chats so one session can run for weeks
token-optimizer-mcp
MCP server and Claude Code plugin that cuts agent token use and keeps a project graph