feat : add headroom proxy for intercept request llm for good compression then save credits

[damnthonyy]

Origin

This PR was initiated from the following prompt:

can you dig deep into the headroom codebase and understand some issues there? also - do a research online for competitors and what in headroom we can enhance to make it better. You can use sub agents.

---

What this PR resolves

Routes Copilot LLM requests through a local proxy (e.g. Headroom) for token compression, monitoring, and quota management — without modifying the core request logic.

---

What changed

New: extensions/copilot/src/platform/networking/common/proxyUtils.ts

• isLLMEndpoint(url) — restricts interception to known Copilot LLM hostnames (api.githubcopilot.com and variants)
• getConfiguredProxyUrl() — reads the COPILOT_PROXY_URL env var
• maybeInterceptUrlThroughProxy() — rewrites the request URL to route through the proxy and injects X-Original-Url / X-Original-Host headers so the proxy knows where to forward

Modified: extensions/copilot/src/platform/endpoint/common/capiClient.ts

• Added ProxyInterceptingFetcherService — wraps the internal fetcher to intercept CAPI requests (standard Copilot chat completions) when COPILOT_PROXY_URL is set and the resolved URL is an LLM endpoint

Modified: extensions/copilot/src/platform/networking/common/networking.ts

• Wired getConfiguredProxyUrl() and maybeInterceptUrlThroughProxy() into networkRequest() for direct URL / BYOK requests
• Extracted inline proxy functions to proxyUtils.ts

Coverage summary

Request path	Intercepted
CAPI (standard Copilot chat)	Yes — via ProxyInterceptingFetcherService
Direct URL / BYOK endpoints	Yes — via networkRequest()
Non-LLM CAPI requests	No — filtered by isLLMEndpoint()

---

How to test

COPILOT_PROXY_URL=http://localhost:4017 ./scripts/code.sh

Make a chat request in Copilot. The POST to api.githubcopilot.com/chat/completions will be intercepted and routed through the proxy.

---

Evidence — real token compression with Headroom

Tested end-to-end with Headroom v0.24.0:

• 2.2k tokens saved in a single session
• $0.0044 lifetime savings (proxy compression only)
• Checkpoint recorded at 14:59 on 11 June 2026
• Status: Healthy — no request failures observed

Enregistrement.de.l.ecran.2026-06-11.a.15.00.13.mov

---

Todo

• Intercept direct URL / BYOK requests
• Add support for CAPI requests (via ProxyInterceptingFetcherService)
• Test end-to-end with real Headroom instance (token compression confirmed)
• Measure token reduction impact and latency overhead at scale
• Handle streaming responses correctly through the proxy
• Fix remaining build issues (@parcel/watcher native module, mermaid types)

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[damnthonyy]

@microsoft-github-policy-service agree