Outline collection pushed; Obsidian note links the map and the collection; Headroom paths recorded
This commit is contained in:
@@ -771,3 +771,24 @@ the earlier test was trying, and it fails today only because LiteLLM does not kn
|
||||
|
||||
**What the person does:** choose A or B. A needs no file edits. B needs the two files above and a
|
||||
container restart, and then one agent's model changed.
|
||||
|
||||
### Headroom, final state after testing every path
|
||||
|
||||
Tested on a throwaway container on a spare port, so the running one was not touched.
|
||||
|
||||
| Configuration tried | Result |
|
||||
|---|---|
|
||||
| default image (`HEADROOM_BACKEND` unset, so `anthropic`) with our router key | `502 proxy_error` |
|
||||
| plus `OPENAI_API_KEY` | `openai/deepseek-v4-flash` → `404 No active credentials for provider: openai`; `deepseek/deepseek-v4-flash` → `502` |
|
||||
| plus `HEADROOM_BACKEND=openai` | `502 proxy_error` |
|
||||
|
||||
Headroom's own `/stats` reports `api_requests: 0` for those calls, so the proxy refuses or fails
|
||||
before it forwards anything. The `404 … provider: openai` is the **router** answering, which means
|
||||
it parsed `openai/` as a provider name and had no key for it — not a headroom misconfiguration.
|
||||
|
||||
What works, and was re-checked: `POST /v1/compress` answers 200 and reports `tokens_before` and
|
||||
`tokens_after`. That is the endpoint `headroom-bridge` uses, and `headroom-ai` is installed on all
|
||||
three machines.
|
||||
|
||||
So: **use the tool path.** The provider path needs someone to work on headroom's own deployment
|
||||
with its logs in front of them; it is not something our software can fix.
|
||||
|
||||
Reference in New Issue
Block a user