Issue connecting to grok model

[machingclee]

I was constantly plagued with the problem from clauded code stating that

API Error: Server error mid-response. The response above may be incomplete

I discussed with grok, back end forth (fixing, trying), and spotted the root cause is an SSE event that may seem valid to grok but invalid to claude-code-proxy (like "event: { ... }" instead of "data: { ... }").

After vibe coding the steps to fix this are as follows, and all my problems with claude code are gone:

1. Clone/use https://github.com/raine/claude-code-proxy (latest release, e.g. v0.1.26).
2. Optionally cherry-pick/merge open PR fix(grok): stabilize prefix cache and long-stream timing #57 (Grok prefix cache / no 120s timeout / reasoning replay) if not already in the release.
3. In src/providers/grok/client.rs:
   • Remove whole-request .timeout(Duration::from_secs(120)).
   • Keep .connect_timeout(10s) + tcp_nodelay.
   • Do NOT enable http2_keep_alive_interval / http2_keep_alive_timeout (they abort during silent Grok reasoning).
4. In src/providers/grok/translate/stream.rs SseDecoder/parse_frame:
   • Comment-only and id-only SSE frames must be ignored (Ok(None)), NOT "Grok SSE frame lacks data" errors.
5. Optional but recommended in src/providers/grok/mod.rs stream loop:
   • While waiting on upstream, every ~5s emit SSE comment ": keepalive\n\n" to Claude so client idle (~15s) doesn't abort.
   • Log mid-stream failures (malformed_sse etc.) even when HTTP status is already 200.
6. cargo test (at least providers::grok), cargo build --release, install binary somewhere early on PATH, restart claude-code-proxy serve.
7. Verify: long high-effort multi-tool Grok session completes without mid-response; proxy.log should not show grok_stream_failed/malformed_sse on normal turns.

Would someone help make an equivalent fix in rust (I am only a beginner level Rustacean and don't want to pollute the codebase)?

Thanks alot.

[raine]

Thanks for taking the time to write all of this up. I just released v0.1.27, which removes the 120-second request timeout and ignores SSE control frames without data. It also handles more of Grok's terminal and control events and logs more detail when a stream fails.

This may help with what you were seeing. Could you give v0.1.27 a try and tell me if it still happens? If it does, any related entries from the proxy log would help me track it down.

[machingclee]

worked flawlessly, thanks.