Skip to content

Commit b267bdf

Browse files
committed
Document Codex Ollama base URL normalization
Update the README and API guide to reflect the shared core behavior for Codex/Ollama base URLs. Root Ollama URLs are now normalized to the OpenAI-compatible /v1 endpoint, so the docs now describe the stable contract that the shared subprocess core enforces.
1 parent c45e036 commit b267bdf

2 files changed

Lines changed: 7 additions & 5 deletions

File tree

README.md

Lines changed: 4 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -188,9 +188,10 @@ The SDK does not infer those flags on its own.
188188
- CLI argument rendering only emits `--model` from a non-empty resolved value
189189

190190
If `ollama_base_url:` is supplied, that endpoint is carried inside the
191-
payload-owned env overrides as `CODEX_OSS_BASE_URL`. The exec and app-server
192-
transports both consume that payload data directly instead of keeping a second
193-
raw base-url path alive downstream.
191+
payload-owned env overrides as `CODEX_OSS_BASE_URL`. Raw Ollama roots are
192+
normalized to the OpenAI-compatible `/v1` base before those env overrides are
193+
emitted. The exec and app-server transports both consume that payload data
194+
directly instead of keeping a second raw base-url path alive downstream.
194195

195196
When the chosen local model is outside Codex's built-in model metadata catalog,
196197
the upstream CLI may warn that it is using fallback metadata. That is an

guides/03-api-guide.md

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -228,8 +228,9 @@ That matters most on the local Ollama route:
228228
- conflicting raw attrs fail during normalization instead of being silently
229229
reinterpreted in the SDK
230230
- a custom Codex/Ollama base URL is carried inside payload env overrides as
231-
`CODEX_OSS_BASE_URL`, so exec and app-server transports can rely on the
232-
payload alone after normalization
231+
`CODEX_OSS_BASE_URL`. Raw Ollama roots are normalized to the
232+
OpenAI-compatible `/v1` base, so exec and app-server transports can rely on
233+
the payload alone after normalization
233234

234235
`connect/2` also accepts `oauth:` for child-auth-aware startup:
235236

0 commit comments

Comments
 (0)