The request-path proxy cut 74% off a live DeepSeek call — see how #1
sudo-ai-git
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
This proxy sits between your agent and any OpenAI-compatible provider, deduping redundant tool results before they are billed.
Live proof (not a benchmark): I routed a tool-heavy request through `mcp-token-saver-proxy` to DeepSeek. The request re-sent a large manifest 5x. The proxy removed 416 of 562 input tokens — 74% of what would have been billed, measured via the `X-Token-Saver-Saved` header, and returned a real completion.
Live endpoint (point any agent `base_url` at it): `https://mcp-token-saver-proxy.fly.dev\`
Reproduce it yourself: `python3 show_savings.py`
Deterministic, no-LLM in the optimization path, streaming, MIT core.
Would love feedback from anyone running long agent loops — what does your stack look like?
All reactions