codex-as-api 0.10.0__tar.gz → 0.10.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/PKG-INFO +31 -177
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/README.md +27 -173
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/RELEASES.md +36 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/pyproject.toml +4 -4
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/scripts/check_package_versions.py +48 -2
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/scripts/check_release_artifacts.py +1 -1
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/__init__.py +1 -1
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/provider.py +9 -3
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/server.py +5 -3
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_package_versions.py +63 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_provider.py +75 -3
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_registry_release.py +16 -16
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_release_artifacts.py +3 -3
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_server.py +57 -1
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_standalone_login_api.py +4 -1
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/.gitignore +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/LICENSE +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/THIRD_PARTY_NOTICES.md +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/config/codex-upstream-contract.json +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/config/o200k_base.tiktoken +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/config/verified-hidden-models.json +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/requirements.txt +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/rust/THIRD_PARTY_NOTICES.md +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/scripts/check_no_static_model_catalog.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/scripts/check_registry_release.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/scripts/generate_rust_third_party_notices.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/scripts/package_rust_release.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/anthropic_adapter.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/auth.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/browser_login.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/codex_config.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/codex_version.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/device_login.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/messages.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/model_capabilities.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/o200k_tokenizer.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/protocol.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/reasoning_signature.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/src/codex_as_api/strict_json.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/__init__.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/conftest.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/fixtures/o200k_base_encode_ordinary.json +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_anthropic_adapter.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_auth.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_browser_login.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_codex_config.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_codex_version.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_device_login.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_login_commit_races.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_login_generation.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_login_http_real_store.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_management_body.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_messages.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_model_capabilities.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_no_static_model_catalog.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_o200k_tokenizer.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_protocol.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_rust_third_party_notices.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_strict_json.py +0 -0
- {codex_as_api-0.10.0 → codex_as_api-0.10.2}/tests/test_ultrafast.py +0 -0
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
Metadata-Version: 2.5
|
|
2
2
|
Name: codex-as-api
|
|
3
|
-
Version: 0.10.
|
|
3
|
+
Version: 0.10.2
|
|
4
4
|
Summary: Use ChatGPT/Codex OAuth as a local OpenAI-compatible API server.
|
|
5
|
-
Project-URL: Homepage, https://github.com/
|
|
6
|
-
Project-URL: Repository, https://github.com/
|
|
7
|
-
Author:
|
|
5
|
+
Project-URL: Homepage, https://github.com/CHOAM-CO/codex-as-api
|
|
6
|
+
Project-URL: Repository, https://github.com/CHOAM-CO/codex-as-api
|
|
7
|
+
Author: CHOAM
|
|
8
8
|
License: Apache-2.0
|
|
9
9
|
License-File: LICENSE
|
|
10
10
|
Keywords: api,chatgpt,codex,llm,oauth,openai
|
|
@@ -35,10 +35,10 @@ Description-Content-Type: text/markdown
|
|
|
35
35
|
|
|
36
36
|
# codex-as-api
|
|
37
37
|
|
|
38
|
-
[](https://github.com/CHOAM-CO/codex-as-api/releases)
|
|
39
39
|
[](https://pypi.org/project/codex-as-api/)
|
|
40
40
|
[](https://www.npmjs.com/package/codex-as-api)
|
|
41
|
-
[](LICENSE)
|
|
42
42
|
|
|
43
43
|
Use ChatGPT / Codex OAuth as a local OpenAI-compatible API server.
|
|
44
44
|
|
|
@@ -100,7 +100,7 @@ codex-as-api
|
|
|
100
100
|
Or from source:
|
|
101
101
|
|
|
102
102
|
```bash
|
|
103
|
-
git clone https://github.com/
|
|
103
|
+
git clone https://github.com/CHOAM-CO/codex-as-api.git
|
|
104
104
|
cd codex-as-api
|
|
105
105
|
pip install -e .
|
|
106
106
|
codex-as-api
|
|
@@ -241,6 +241,16 @@ admin bearer and is not a replacement for authentication or network controls.
|
|
|
241
241
|
Each credential store must be owned by one server process, not shared between
|
|
242
242
|
independent workers. Choose separate store paths for separate instances.
|
|
243
243
|
|
|
244
|
+
### Note on OpenAI "Sign in with ChatGPT" (SIWC) preview
|
|
245
|
+
|
|
246
|
+
OpenAI's public [Sign in with ChatGPT (`chatgpt.tokens.use.direct`)](https://developers.openai.com/cookbook/articles/sign-in-with-chatgpt)
|
|
247
|
+
preview route (`https://api.openai.com/v1/responses` with dynamic client registration)
|
|
248
|
+
is intentionally kept on hold and not used as the default transport:
|
|
249
|
+
|
|
250
|
+
- Function tools on the SIWC preview route must be wrapped inside a `namespace` tool object rather than sent as top-level function definitions.
|
|
251
|
+
- Remote compaction (`/responses/compact`), `Responses Lite`, hosted `image_generation`, device-code login, and ChatGPT Live / Realtime WebRTC voice calls are not supported on the SIWC preview route.
|
|
252
|
+
- Reliable prompt cache hits on SIWC currently require reusing a persistent WebSocket (`wss://api.openai.com/v1/responses`) rather than stateless HTTP `POST /v1/responses` requests.
|
|
253
|
+
|
|
244
254
|
## Configuration
|
|
245
255
|
|
|
246
256
|
Environment variables (Python, Rust, and TypeScript):
|
|
@@ -288,7 +298,7 @@ These three upstream-hidden models returned real responses on October 24, 2026,
|
|
|
288
298
|
|
|
289
299
|
A small authenticated smoke test also verified function-tool calling plus tool-result continuation for all three models, and executed a simple generated JavaScript function against four inputs per model. This is a basic functionality check, not a coding-quality benchmark.
|
|
290
300
|
|
|
291
|
-
**These are hidden models. Upstream support can stop at any time, and availability may differ by account. If a model stops responding, [open an issue](https://github.com/
|
|
301
|
+
**These are hidden models. Upstream support can stop at any time, and availability may differ by account. If a model stops responding, [open an issue](https://github.com/CHOAM-CO/codex-as-api/issues) with its model ID and a sanitized error; we will remove the discontinued model from this catalog.** Do not include access tokens or other credentials in the issue.
|
|
292
302
|
|
|
293
303
|
The proxy exposes these verified exceptions as selectable models and prefers their current live capability metadata. Bundled capability records keep only these three IDs available when a successful live catalog omits them; they do not bypass OAuth, upstream rejection, or a failed/malformed catalog fetch. A successful empty upstream catalog can therefore list these experimental entries, while health/default selection still fails until an ordinary upstream default is available. The proxy never selects a hidden experimental model as its automatic default.
|
|
294
304
|
|
|
@@ -451,6 +461,10 @@ null, or server-tool caller variants still fail before transport.
|
|
|
451
461
|
|
|
452
462
|
### Identifier and cache behavior
|
|
453
463
|
|
|
464
|
+
For chat requests, initial system messages form `instructions`. System
|
|
465
|
+
messages added after conversation history starts stay at their original positions
|
|
466
|
+
as developer messages, so appending runtime notes preserves the earlier history prefix.
|
|
467
|
+
|
|
454
468
|
The two compatibility facades intentionally use the similarly named fields for
|
|
455
469
|
different jobs:
|
|
456
470
|
|
|
@@ -967,10 +981,12 @@ The provider handles:
|
|
|
967
981
|
## Release & package publishing
|
|
968
982
|
|
|
969
983
|
The `Release` GitHub Actions workflow runs on a pushed `v*` tag. It rejects a
|
|
970
|
-
tag that does not exactly match every Python, npm, and Rust version surface
|
|
984
|
+
tag that does not exactly match every Python, npm, and Rust version surface or
|
|
985
|
+
lacks a matching `## v<version>` section in [RELEASES.md](RELEASES.md),
|
|
971
986
|
runs the complete cross-runtime gate, and then publishes PyPI, npmjs, the
|
|
972
987
|
scoped GitHub npm package, deterministic platform Rust ZIP archives, and one
|
|
973
|
-
GitHub Release
|
|
988
|
+
GitHub Release whose body is populated from that `RELEASES.md` section. Each
|
|
989
|
+
Rust archive contains the platform binary, `LICENSE`, and
|
|
974
990
|
the generated `THIRD_PARTY_NOTICES.md` covering both bundled tokenizer data and
|
|
975
991
|
the locked normal dependency graph across supported release targets.
|
|
976
992
|
|
|
@@ -980,10 +996,10 @@ registry setup is still required:
|
|
|
980
996
|
|
|
981
997
|
- Create GitHub environments named `pypi` and `npm`; add deployment protection
|
|
982
998
|
rules if desired.
|
|
983
|
-
- On PyPI, add a trusted publisher for project `codex-as-api`, owner `
|
|
999
|
+
- On PyPI, add a trusted publisher for project `codex-as-api`, owner `CHOAM-CO`,
|
|
984
1000
|
repository `codex-as-api`, workflow `release.yml`, environment `pypi`.
|
|
985
1001
|
- On npmjs, add a GitHub Actions trusted publisher for package `codex-as-api`,
|
|
986
|
-
owner `
|
|
1002
|
+
owner `CHOAM-CO`, repository `codex-as-api`, workflow `release.yml`,
|
|
987
1003
|
environment `npm`, with `npm publish` allowed.
|
|
988
1004
|
- GitHub Packages and GitHub Releases use the job-scoped automatic
|
|
989
1005
|
`GITHUB_TOKEN`; repository policy must allow the workflow's declared
|
|
@@ -992,9 +1008,9 @@ registry setup is still required:
|
|
|
992
1008
|
After bumping every version and passing the local gate, push the matching tag:
|
|
993
1009
|
|
|
994
1010
|
```bash
|
|
995
|
-
python scripts/check_package_versions.py --tag v0.10.
|
|
996
|
-
git tag v0.10.
|
|
997
|
-
git push origin v0.10.
|
|
1011
|
+
python scripts/check_package_versions.py --tag v0.10.2
|
|
1012
|
+
git tag v0.10.2
|
|
1013
|
+
git push origin v0.10.2
|
|
998
1014
|
```
|
|
999
1015
|
|
|
1000
1016
|
## Tests
|
|
@@ -1022,168 +1038,6 @@ npm install
|
|
|
1022
1038
|
npm test
|
|
1023
1039
|
```
|
|
1024
1040
|
|
|
1025
|
-
|
|
1026
|
-
## Release Notes
|
|
1027
|
-
|
|
1028
|
-
### v0.10.0
|
|
1029
|
-
|
|
1030
|
-
- Sign in directly with ordinary browser OAuth; no Codex installation or existing
|
|
1031
|
-
Codex credentials are required. Device-code login is an explicit remote option.
|
|
1032
|
-
- Add independent admin-key-protected login start/status/cancel endpoints and a
|
|
1033
|
-
private application-owned credential store across Python, TypeScript and Rust.
|
|
1034
|
-
- Upgrading requires a new codex-as-api login; existing Codex credentials remain
|
|
1035
|
-
untouched. See [Standalone login](#standalone-login) and [full release notes](RELEASES.md).
|
|
1036
|
-
|
|
1037
|
-
### v0.8.0
|
|
1038
|
-
|
|
1039
|
-
- Add ChatGPT Live realtime voice to the TypeScript runtime: `POST
|
|
1040
|
-
/v1/realtime/calls` creates an OAuth-authorized call from a client-owned WebRTC
|
|
1041
|
-
SDP offer and returns the SDP answer, and `WS /v1/realtime/sideband/{call_id}`
|
|
1042
|
-
relays the OpenAI realtime event sideband for headless clients. Audio stays on
|
|
1043
|
-
the client's media plane.
|
|
1044
|
-
- Mirror the pinned Codex call contract exactly: one identity in
|
|
1045
|
-
`x-session-id`/`session-id`/`thread-id`, the `openai-alpha: quicksilver=v2`
|
|
1046
|
-
handshake, `intent=quicksilver&architecture=avas`, and a session payload that
|
|
1047
|
-
never chooses its own voice-session id.
|
|
1048
|
-
- Document realtime as a TypeScript-only capability; Python and Rust remain
|
|
1049
|
-
HTTP-only in this release instead of claiming cross-runtime parity.
|
|
1050
|
-
|
|
1051
|
-
### v0.7.0
|
|
1052
|
-
|
|
1053
|
-
- Replace bundled model capability tables with a fresh authenticated Codex
|
|
1054
|
-
catalog and add `GET /v1/models` across all three runtimes.
|
|
1055
|
-
- Remove stale/unknown-model fallbacks, fabricated Anthropic signatures, silent
|
|
1056
|
-
request downgrades, and malformed upstream recovery; return typed failures.
|
|
1057
|
-
- Reject hosted Anthropic WebSearch instead of translating incomplete OpenAI
|
|
1058
|
-
search lifecycle data, and require authoritative completion and final usage.
|
|
1059
|
-
- Keep immediate Claude Code `message_start` delivery without inventing token
|
|
1060
|
-
counts; final Anthropic fields are emitted only from authoritative provider data.
|
|
1061
|
-
- Scope continuation history and catalog state to the authenticated account,
|
|
1062
|
-
validate before streaming, and make `/health` report actual readiness.
|
|
1063
|
-
- Require the official nested `tokens` auth layout, remove root-level and
|
|
1064
|
-
noncanonical auth aliases, and reject external-host `chatgptAuthTokens`.
|
|
1065
|
-
- Pin the validated private contract to Codex `0.153.3` and publish the exact
|
|
1066
|
-
tested artifacts to PyPI, npmjs, GitHub Packages, and GitHub Releases.
|
|
1067
|
-
- Install Python server dependencies with the base package so the default
|
|
1068
|
-
`pip install codex-as-api` command produces a runnable server.
|
|
1069
|
-
|
|
1070
|
-
### v0.6.5
|
|
1071
|
-
|
|
1072
|
-
- Align GPT-5.6 alias, Sol, Terra, and Luna context limits with Codex `0.147.0`
|
|
1073
|
-
at 272,000 tokens and derive the 244,800-token automatic compact threshold.
|
|
1074
|
-
- Pin the verified upstream request, Responses Lite, header, and model contract
|
|
1075
|
-
to `openai/codex` commit
|
|
1076
|
-
`be6e8eac029b183056b7e4402879f15d2c85f61b`; track its new fractional SSE
|
|
1077
|
-
rollout-budget field without claiming an unapproved facade mapping.
|
|
1078
|
-
- Refresh OAuth credentials five minutes before expiry, coalesce concurrent
|
|
1079
|
-
refreshes, reload changed credentials after 401, and preserve upstream HTTP
|
|
1080
|
-
status codes across all three servers.
|
|
1081
|
-
- Replace startup npm version discovery with deterministic compatibility and
|
|
1082
|
-
package-version identity in the Codex-style `User-Agent`.
|
|
1083
|
-
- Add full cross-runtime CI plus tag-gated OIDC publishing to PyPI and npmjs,
|
|
1084
|
-
`GITHUB_TOKEN` publishing to GitHub Packages, and GitHub Release assets.
|
|
1085
|
-
|
|
1086
|
-
### v0.6.4
|
|
1087
|
-
|
|
1088
|
-
- Remove process-global generated Codex session/thread identities and require
|
|
1089
|
-
explicit session identity when Codex metadata mode is enabled.
|
|
1090
|
-
- Resolve Chat cache affinity as explicit `prompt_cache_key`, then
|
|
1091
|
-
`client_metadata.session_id`, then omission.
|
|
1092
|
-
- Derive Claude Code cache affinity from `x-claude-code-session-id` without
|
|
1093
|
-
fabricating Codex metadata; keep normal Anthropic Messages stateless.
|
|
1094
|
-
- Validate and strip Anthropic `cache_control` compatibility hints without
|
|
1095
|
-
claiming unavailable breakpoint or TTL semantics.
|
|
1096
|
-
- Update current compatibility evidence to Codex `0.145.0`, Codex `main` at
|
|
1097
|
-
`bd2de422aa287b97b06ca6425a10935bcf1b3731`, and Claude Code `2.1.220`.
|
|
1098
|
-
|
|
1099
|
-
### v0.6.3
|
|
1100
|
-
|
|
1101
|
-
- Accept Claude Code auxiliary requests that combine ambient `output_config.effort` with call-level `thinking.disabled`.
|
|
1102
|
-
- Give explicit disabled thinking precedence and send Codex reasoning effort `none` instead of returning HTTP 400.
|
|
1103
|
-
- Document and test `CODEX_AS_API_RESPONSES_LITE=off` for Claude Code hosted WebSearch on GPT-5.6; WebFetch needs no Responses Lite override.
|
|
1104
|
-
- Preserve fail-loudly validation for invalid effort values and unsupported `output_config` fields.
|
|
1105
|
-
- Add Python, TypeScript, and Rust adapter regressions, streamed `/v1/messages` coverage, and a `/count_tokens` regression for the WebSearch/WebFetch request shape.
|
|
1106
|
-
|
|
1107
|
-
### v0.6.2
|
|
1108
|
-
|
|
1109
|
-
- Fix `/v1/messages/count_tokens` overcounting that could trigger Claude Code autocompaction on every turn.
|
|
1110
|
-
- Count normalized model-visible messages, tool calls, reasoning, tool-result metadata, and tool schemas once; remove the second full raw-request-body addition.
|
|
1111
|
-
- Port official `tiktoken` `o200k_base` ordinary encoding into Python, TypeScript, and Rust without adding a `tiktoken` package dependency or runtime rank download.
|
|
1112
|
-
- Match the upstream Unicode text split and BPE merge ranks last checked on 2026-07-14 against `tiktoken` 0.13.0 at `08a5f3b`.
|
|
1113
|
-
- Keep image input cost separate and count inline base64 images once rather than adding both a fixed image estimate and the base64 request bytes.
|
|
1114
|
-
- Add matching Python, TypeScript, and Rust endpoint regressions: a 4,000-byte ASCII message now returns `1,012` instead of roughly `8,000`, and non-model control fields do not change the count.
|
|
1115
|
-
- Validate the release with Claude Code `2.1.209` and real Codex OAuth chat.
|
|
1116
|
-
|
|
1117
|
-
### v0.6.1
|
|
1118
|
-
|
|
1119
|
-
- Re-audit official Codex `0.144.4` and `main` at `393f64565ab46f09d99ca4d9bd973537e72a114b`, plus Claude Code `2.1.208`, before publishing.
|
|
1120
|
-
- Validate Claude Code `2.1.208` with real Codex OAuth chat and tool loops, and document the official custom GPT picker entry alongside Fable, Opus, Sonnet, and Haiku.
|
|
1121
|
-
- Route bundled GPT model IDs from Anthropic requests to the selected Codex backend model while retaining configured fallback behavior for built-in Claude model names.
|
|
1122
|
-
- Map `output_config.effort` and Fast Mode to Codex effort and priority service tier, preserve URL images and tool-error state, and handle current context/tool beta fields without silent loss.
|
|
1123
|
-
- Keep Messages, token counting, and remote compaction aligned on effective model limits and fail-loudly validation for output formats and image sources.
|
|
1124
|
-
- Keep Python streaming work off the ASGI event loop and make Rust Anthropic SSE truly incremental so concurrent Claude Code requests are not stalled or buffered.
|
|
1125
|
-
- Preserve Rust streaming authentication, rate-limit, and overload errors after OAuth refresh instead of collapsing them to generic HTTP 500 failures.
|
|
1126
|
-
- Restrict the Python source distribution to release inputs so local agent state and unrelated runtime sources cannot enter PyPI artifacts.
|
|
1127
|
-
|
|
1128
|
-
### v0.6.0
|
|
1129
|
-
|
|
1130
|
-
- Add GPT-5.6 public alias plus Sol, Terra, and Luna capability metadata, model defaults, and Codex context-window behavior.
|
|
1131
|
-
- Refresh bundled context and default-effort metadata for current GPT-5.5, GPT-5.4, GPT-5.4 Mini, and GPT-5.2 catalog entries.
|
|
1132
|
-
- Update Responses Lite request bodies and headers across chat, compact, and inspection paths; Lite image generation is rejected explicitly because this proxy has no standalone image-tool executor.
|
|
1133
|
-
- Add `max`, Codex-compatible `ultra` to `max` wire conversion, future model-defined efforts, and `model_reasoning_effort` config support.
|
|
1134
|
-
- Add GPT-5.6 reasoning effort/context, `standard` mode compatibility, capability-gated image `original` detail, standard verbosity, cache accounting, and real backend `response_id` support across Python, TypeScript, and Rust.
|
|
1135
|
-
- Translate known `previous_response_id` values into bounded local full-history replay; reject Pro and non-null public cache policy/breakpoints or `safety_identifier` because the private Codex request contract has no equivalent field.
|
|
1136
|
-
- Reject hosted Multi-agent and Programmatic Tool Calling on the Chat facade until their native agent/program item lifecycle can be preserved.
|
|
1137
|
-
|
|
1138
|
-
### v0.5.2
|
|
1139
|
-
|
|
1140
|
-
- Support latest Codex root-level OAuth token files while keeping PAT-only, agent-identity-only, and Bedrock-only auth files explicitly unsupported.
|
|
1141
|
-
- Add shared model capability gating for Responses Lite, parallel tool calls, verbosity, and service-tier behavior across Python, TypeScript, and Rust.
|
|
1142
|
-
- Preserve encrypted reasoning state via top-level `reasoning.encrypted_content` include and add Codex metadata forwarding controls.
|
|
1143
|
-
|
|
1144
|
-
### v0.5.1
|
|
1145
|
-
|
|
1146
|
-
- Add official Codex CLI `originator` and versioned `User-Agent` headers for ChatGPT/Codex OAuth requests.
|
|
1147
|
-
- Resolve the latest `@openai/codex` version from npm at server startup, with `CODEX_AS_API_CODEX_CLI_VERSION` as an explicit override.
|
|
1148
|
-
|
|
1149
|
-
### v0.5.0
|
|
1150
|
-
|
|
1151
|
-
- Preserve Claude Code server-tool history (`server_tool_use`, `web_search_tool_result`, MCP/advisor-like result blocks) as backend context instead of dropping it on the next turn.
|
|
1152
|
-
- Support Anthropic `output_format` structured outputs by mapping JSON schema/object formats to OpenAI Responses `text.format`.
|
|
1153
|
-
- Preserve `document` and `search_result` content blocks inside tool results, keep Python streaming defaults aligned, and accept unsuffixed `web_search` server-tool types across Python, TypeScript, and Rust.
|
|
1154
|
-
|
|
1155
|
-
### v0.4.0
|
|
1156
|
-
|
|
1157
|
-
- Add Claude Code-compatible Anthropic hosted web search support by mapping `web_search_*` server tools to OpenAI Responses `web_search`.
|
|
1158
|
-
- Return `server_tool_use` and `web_search_tool_result` blocks so Claude Code can parse web search results reliably.
|
|
1159
|
-
- Prepare TypeScript package publishing to npmjs and GitHub Packages via GitHub Actions.
|
|
1160
|
-
|
|
1161
|
-
### v0.3.3
|
|
1162
|
-
|
|
1163
|
-
- Stop forwarding client `max_tokens` as Codex `max_output_tokens`, restoring Claude Code compatibility with the Codex OAuth backend.
|
|
1164
|
-
- Add Python, TypeScript, and Rust regression tests for the provider payload.
|
|
1165
|
-
|
|
1166
|
-
### v0.3.2
|
|
1167
|
-
|
|
1168
|
-
- Restore immediate Anthropic streaming so Claude Code receives events without waiting for the backend response to finish.
|
|
1169
|
-
- Use conservative local token estimates for `/v1/messages/count_tokens`; Codex OAuth has no count-only backend endpoint.
|
|
1170
|
-
- Keep real final streaming usage metadata in `message_delta`.
|
|
1171
|
-
|
|
1172
|
-
### v0.3.1
|
|
1173
|
-
|
|
1174
|
-
- Attempted real backend token counting for `/v1/messages/count_tokens` with `max_output_tokens: 0`; this is superseded by v0.3.2 because Codex OAuth rejects count-only requests.
|
|
1175
|
-
- Forward converted Anthropic tools, tool choice, stop sequences, and thinking/reasoning settings during token-count requests.
|
|
1176
|
-
- Propagate cumulative Anthropic streaming usage, including cache accounting, server tool use, and service tier metadata when available.
|
|
1177
|
-
- Pass `max_output_tokens` through provider requests across Python, TypeScript, and Rust.
|
|
1178
|
-
|
|
1179
|
-
### v0.3.0
|
|
1180
|
-
|
|
1181
|
-
- Read Codex CLI config from `CODEX_HOME` / `~/.codex/config.toml` across Python, TypeScript, and Rust.
|
|
1182
|
-
- Use the configured Codex backend model while preserving Anthropic client model names in `/v1/messages` responses.
|
|
1183
|
-
- Expose `context_window` and `auto_compact_token_limit` through `/health` and `/v1/messages/count_tokens`.
|
|
1184
|
-
- Add Anthropic-compatible `/v1/messages/count_tokens` and `/v1/messages/compact`.
|
|
1185
|
-
- Map context-window failures to Anthropic-style `400 invalid_request_error` responses and stream error events.
|
|
1186
|
-
|
|
1187
1041
|
## License
|
|
1188
1042
|
|
|
1189
1043
|
Apache License 2.0 — derived from [OpenAI Codex CLI](https://github.com/openai/codex) (Apache-2.0, Copyright 2025 OpenAI).
|
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
# codex-as-api
|
|
2
2
|
|
|
3
|
-
[](https://github.com/CHOAM-CO/codex-as-api/releases)
|
|
4
4
|
[](https://pypi.org/project/codex-as-api/)
|
|
5
5
|
[](https://www.npmjs.com/package/codex-as-api)
|
|
6
|
-
[](LICENSE)
|
|
7
7
|
|
|
8
8
|
Use ChatGPT / Codex OAuth as a local OpenAI-compatible API server.
|
|
9
9
|
|
|
@@ -65,7 +65,7 @@ codex-as-api
|
|
|
65
65
|
Or from source:
|
|
66
66
|
|
|
67
67
|
```bash
|
|
68
|
-
git clone https://github.com/
|
|
68
|
+
git clone https://github.com/CHOAM-CO/codex-as-api.git
|
|
69
69
|
cd codex-as-api
|
|
70
70
|
pip install -e .
|
|
71
71
|
codex-as-api
|
|
@@ -206,6 +206,16 @@ admin bearer and is not a replacement for authentication or network controls.
|
|
|
206
206
|
Each credential store must be owned by one server process, not shared between
|
|
207
207
|
independent workers. Choose separate store paths for separate instances.
|
|
208
208
|
|
|
209
|
+
### Note on OpenAI "Sign in with ChatGPT" (SIWC) preview
|
|
210
|
+
|
|
211
|
+
OpenAI's public [Sign in with ChatGPT (`chatgpt.tokens.use.direct`)](https://developers.openai.com/cookbook/articles/sign-in-with-chatgpt)
|
|
212
|
+
preview route (`https://api.openai.com/v1/responses` with dynamic client registration)
|
|
213
|
+
is intentionally kept on hold and not used as the default transport:
|
|
214
|
+
|
|
215
|
+
- Function tools on the SIWC preview route must be wrapped inside a `namespace` tool object rather than sent as top-level function definitions.
|
|
216
|
+
- Remote compaction (`/responses/compact`), `Responses Lite`, hosted `image_generation`, device-code login, and ChatGPT Live / Realtime WebRTC voice calls are not supported on the SIWC preview route.
|
|
217
|
+
- Reliable prompt cache hits on SIWC currently require reusing a persistent WebSocket (`wss://api.openai.com/v1/responses`) rather than stateless HTTP `POST /v1/responses` requests.
|
|
218
|
+
|
|
209
219
|
## Configuration
|
|
210
220
|
|
|
211
221
|
Environment variables (Python, Rust, and TypeScript):
|
|
@@ -253,7 +263,7 @@ These three upstream-hidden models returned real responses on October 24, 2026,
|
|
|
253
263
|
|
|
254
264
|
A small authenticated smoke test also verified function-tool calling plus tool-result continuation for all three models, and executed a simple generated JavaScript function against four inputs per model. This is a basic functionality check, not a coding-quality benchmark.
|
|
255
265
|
|
|
256
|
-
**These are hidden models. Upstream support can stop at any time, and availability may differ by account. If a model stops responding, [open an issue](https://github.com/
|
|
266
|
+
**These are hidden models. Upstream support can stop at any time, and availability may differ by account. If a model stops responding, [open an issue](https://github.com/CHOAM-CO/codex-as-api/issues) with its model ID and a sanitized error; we will remove the discontinued model from this catalog.** Do not include access tokens or other credentials in the issue.
|
|
257
267
|
|
|
258
268
|
The proxy exposes these verified exceptions as selectable models and prefers their current live capability metadata. Bundled capability records keep only these three IDs available when a successful live catalog omits them; they do not bypass OAuth, upstream rejection, or a failed/malformed catalog fetch. A successful empty upstream catalog can therefore list these experimental entries, while health/default selection still fails until an ordinary upstream default is available. The proxy never selects a hidden experimental model as its automatic default.
|
|
259
269
|
|
|
@@ -416,6 +426,10 @@ null, or server-tool caller variants still fail before transport.
|
|
|
416
426
|
|
|
417
427
|
### Identifier and cache behavior
|
|
418
428
|
|
|
429
|
+
For chat requests, initial system messages form `instructions`. System
|
|
430
|
+
messages added after conversation history starts stay at their original positions
|
|
431
|
+
as developer messages, so appending runtime notes preserves the earlier history prefix.
|
|
432
|
+
|
|
419
433
|
The two compatibility facades intentionally use the similarly named fields for
|
|
420
434
|
different jobs:
|
|
421
435
|
|
|
@@ -932,10 +946,12 @@ The provider handles:
|
|
|
932
946
|
## Release & package publishing
|
|
933
947
|
|
|
934
948
|
The `Release` GitHub Actions workflow runs on a pushed `v*` tag. It rejects a
|
|
935
|
-
tag that does not exactly match every Python, npm, and Rust version surface
|
|
949
|
+
tag that does not exactly match every Python, npm, and Rust version surface or
|
|
950
|
+
lacks a matching `## v<version>` section in [RELEASES.md](RELEASES.md),
|
|
936
951
|
runs the complete cross-runtime gate, and then publishes PyPI, npmjs, the
|
|
937
952
|
scoped GitHub npm package, deterministic platform Rust ZIP archives, and one
|
|
938
|
-
GitHub Release
|
|
953
|
+
GitHub Release whose body is populated from that `RELEASES.md` section. Each
|
|
954
|
+
Rust archive contains the platform binary, `LICENSE`, and
|
|
939
955
|
the generated `THIRD_PARTY_NOTICES.md` covering both bundled tokenizer data and
|
|
940
956
|
the locked normal dependency graph across supported release targets.
|
|
941
957
|
|
|
@@ -945,10 +961,10 @@ registry setup is still required:
|
|
|
945
961
|
|
|
946
962
|
- Create GitHub environments named `pypi` and `npm`; add deployment protection
|
|
947
963
|
rules if desired.
|
|
948
|
-
- On PyPI, add a trusted publisher for project `codex-as-api`, owner `
|
|
964
|
+
- On PyPI, add a trusted publisher for project `codex-as-api`, owner `CHOAM-CO`,
|
|
949
965
|
repository `codex-as-api`, workflow `release.yml`, environment `pypi`.
|
|
950
966
|
- On npmjs, add a GitHub Actions trusted publisher for package `codex-as-api`,
|
|
951
|
-
owner `
|
|
967
|
+
owner `CHOAM-CO`, repository `codex-as-api`, workflow `release.yml`,
|
|
952
968
|
environment `npm`, with `npm publish` allowed.
|
|
953
969
|
- GitHub Packages and GitHub Releases use the job-scoped automatic
|
|
954
970
|
`GITHUB_TOKEN`; repository policy must allow the workflow's declared
|
|
@@ -957,9 +973,9 @@ registry setup is still required:
|
|
|
957
973
|
After bumping every version and passing the local gate, push the matching tag:
|
|
958
974
|
|
|
959
975
|
```bash
|
|
960
|
-
python scripts/check_package_versions.py --tag v0.10.
|
|
961
|
-
git tag v0.10.
|
|
962
|
-
git push origin v0.10.
|
|
976
|
+
python scripts/check_package_versions.py --tag v0.10.2
|
|
977
|
+
git tag v0.10.2
|
|
978
|
+
git push origin v0.10.2
|
|
963
979
|
```
|
|
964
980
|
|
|
965
981
|
## Tests
|
|
@@ -987,168 +1003,6 @@ npm install
|
|
|
987
1003
|
npm test
|
|
988
1004
|
```
|
|
989
1005
|
|
|
990
|
-
|
|
991
|
-
## Release Notes
|
|
992
|
-
|
|
993
|
-
### v0.10.0
|
|
994
|
-
|
|
995
|
-
- Sign in directly with ordinary browser OAuth; no Codex installation or existing
|
|
996
|
-
Codex credentials are required. Device-code login is an explicit remote option.
|
|
997
|
-
- Add independent admin-key-protected login start/status/cancel endpoints and a
|
|
998
|
-
private application-owned credential store across Python, TypeScript and Rust.
|
|
999
|
-
- Upgrading requires a new codex-as-api login; existing Codex credentials remain
|
|
1000
|
-
untouched. See [Standalone login](#standalone-login) and [full release notes](RELEASES.md).
|
|
1001
|
-
|
|
1002
|
-
### v0.8.0
|
|
1003
|
-
|
|
1004
|
-
- Add ChatGPT Live realtime voice to the TypeScript runtime: `POST
|
|
1005
|
-
/v1/realtime/calls` creates an OAuth-authorized call from a client-owned WebRTC
|
|
1006
|
-
SDP offer and returns the SDP answer, and `WS /v1/realtime/sideband/{call_id}`
|
|
1007
|
-
relays the OpenAI realtime event sideband for headless clients. Audio stays on
|
|
1008
|
-
the client's media plane.
|
|
1009
|
-
- Mirror the pinned Codex call contract exactly: one identity in
|
|
1010
|
-
`x-session-id`/`session-id`/`thread-id`, the `openai-alpha: quicksilver=v2`
|
|
1011
|
-
handshake, `intent=quicksilver&architecture=avas`, and a session payload that
|
|
1012
|
-
never chooses its own voice-session id.
|
|
1013
|
-
- Document realtime as a TypeScript-only capability; Python and Rust remain
|
|
1014
|
-
HTTP-only in this release instead of claiming cross-runtime parity.
|
|
1015
|
-
|
|
1016
|
-
### v0.7.0
|
|
1017
|
-
|
|
1018
|
-
- Replace bundled model capability tables with a fresh authenticated Codex
|
|
1019
|
-
catalog and add `GET /v1/models` across all three runtimes.
|
|
1020
|
-
- Remove stale/unknown-model fallbacks, fabricated Anthropic signatures, silent
|
|
1021
|
-
request downgrades, and malformed upstream recovery; return typed failures.
|
|
1022
|
-
- Reject hosted Anthropic WebSearch instead of translating incomplete OpenAI
|
|
1023
|
-
search lifecycle data, and require authoritative completion and final usage.
|
|
1024
|
-
- Keep immediate Claude Code `message_start` delivery without inventing token
|
|
1025
|
-
counts; final Anthropic fields are emitted only from authoritative provider data.
|
|
1026
|
-
- Scope continuation history and catalog state to the authenticated account,
|
|
1027
|
-
validate before streaming, and make `/health` report actual readiness.
|
|
1028
|
-
- Require the official nested `tokens` auth layout, remove root-level and
|
|
1029
|
-
noncanonical auth aliases, and reject external-host `chatgptAuthTokens`.
|
|
1030
|
-
- Pin the validated private contract to Codex `0.153.3` and publish the exact
|
|
1031
|
-
tested artifacts to PyPI, npmjs, GitHub Packages, and GitHub Releases.
|
|
1032
|
-
- Install Python server dependencies with the base package so the default
|
|
1033
|
-
`pip install codex-as-api` command produces a runnable server.
|
|
1034
|
-
|
|
1035
|
-
### v0.6.5
|
|
1036
|
-
|
|
1037
|
-
- Align GPT-5.6 alias, Sol, Terra, and Luna context limits with Codex `0.147.0`
|
|
1038
|
-
at 272,000 tokens and derive the 244,800-token automatic compact threshold.
|
|
1039
|
-
- Pin the verified upstream request, Responses Lite, header, and model contract
|
|
1040
|
-
to `openai/codex` commit
|
|
1041
|
-
`be6e8eac029b183056b7e4402879f15d2c85f61b`; track its new fractional SSE
|
|
1042
|
-
rollout-budget field without claiming an unapproved facade mapping.
|
|
1043
|
-
- Refresh OAuth credentials five minutes before expiry, coalesce concurrent
|
|
1044
|
-
refreshes, reload changed credentials after 401, and preserve upstream HTTP
|
|
1045
|
-
status codes across all three servers.
|
|
1046
|
-
- Replace startup npm version discovery with deterministic compatibility and
|
|
1047
|
-
package-version identity in the Codex-style `User-Agent`.
|
|
1048
|
-
- Add full cross-runtime CI plus tag-gated OIDC publishing to PyPI and npmjs,
|
|
1049
|
-
`GITHUB_TOKEN` publishing to GitHub Packages, and GitHub Release assets.
|
|
1050
|
-
|
|
1051
|
-
### v0.6.4
|
|
1052
|
-
|
|
1053
|
-
- Remove process-global generated Codex session/thread identities and require
|
|
1054
|
-
explicit session identity when Codex metadata mode is enabled.
|
|
1055
|
-
- Resolve Chat cache affinity as explicit `prompt_cache_key`, then
|
|
1056
|
-
`client_metadata.session_id`, then omission.
|
|
1057
|
-
- Derive Claude Code cache affinity from `x-claude-code-session-id` without
|
|
1058
|
-
fabricating Codex metadata; keep normal Anthropic Messages stateless.
|
|
1059
|
-
- Validate and strip Anthropic `cache_control` compatibility hints without
|
|
1060
|
-
claiming unavailable breakpoint or TTL semantics.
|
|
1061
|
-
- Update current compatibility evidence to Codex `0.145.0`, Codex `main` at
|
|
1062
|
-
`bd2de422aa287b97b06ca6425a10935bcf1b3731`, and Claude Code `2.1.220`.
|
|
1063
|
-
|
|
1064
|
-
### v0.6.3
|
|
1065
|
-
|
|
1066
|
-
- Accept Claude Code auxiliary requests that combine ambient `output_config.effort` with call-level `thinking.disabled`.
|
|
1067
|
-
- Give explicit disabled thinking precedence and send Codex reasoning effort `none` instead of returning HTTP 400.
|
|
1068
|
-
- Document and test `CODEX_AS_API_RESPONSES_LITE=off` for Claude Code hosted WebSearch on GPT-5.6; WebFetch needs no Responses Lite override.
|
|
1069
|
-
- Preserve fail-loudly validation for invalid effort values and unsupported `output_config` fields.
|
|
1070
|
-
- Add Python, TypeScript, and Rust adapter regressions, streamed `/v1/messages` coverage, and a `/count_tokens` regression for the WebSearch/WebFetch request shape.
|
|
1071
|
-
|
|
1072
|
-
### v0.6.2
|
|
1073
|
-
|
|
1074
|
-
- Fix `/v1/messages/count_tokens` overcounting that could trigger Claude Code autocompaction on every turn.
|
|
1075
|
-
- Count normalized model-visible messages, tool calls, reasoning, tool-result metadata, and tool schemas once; remove the second full raw-request-body addition.
|
|
1076
|
-
- Port official `tiktoken` `o200k_base` ordinary encoding into Python, TypeScript, and Rust without adding a `tiktoken` package dependency or runtime rank download.
|
|
1077
|
-
- Match the upstream Unicode text split and BPE merge ranks last checked on 2026-07-14 against `tiktoken` 0.13.0 at `08a5f3b`.
|
|
1078
|
-
- Keep image input cost separate and count inline base64 images once rather than adding both a fixed image estimate and the base64 request bytes.
|
|
1079
|
-
- Add matching Python, TypeScript, and Rust endpoint regressions: a 4,000-byte ASCII message now returns `1,012` instead of roughly `8,000`, and non-model control fields do not change the count.
|
|
1080
|
-
- Validate the release with Claude Code `2.1.209` and real Codex OAuth chat.
|
|
1081
|
-
|
|
1082
|
-
### v0.6.1
|
|
1083
|
-
|
|
1084
|
-
- Re-audit official Codex `0.144.4` and `main` at `393f64565ab46f09d99ca4d9bd973537e72a114b`, plus Claude Code `2.1.208`, before publishing.
|
|
1085
|
-
- Validate Claude Code `2.1.208` with real Codex OAuth chat and tool loops, and document the official custom GPT picker entry alongside Fable, Opus, Sonnet, and Haiku.
|
|
1086
|
-
- Route bundled GPT model IDs from Anthropic requests to the selected Codex backend model while retaining configured fallback behavior for built-in Claude model names.
|
|
1087
|
-
- Map `output_config.effort` and Fast Mode to Codex effort and priority service tier, preserve URL images and tool-error state, and handle current context/tool beta fields without silent loss.
|
|
1088
|
-
- Keep Messages, token counting, and remote compaction aligned on effective model limits and fail-loudly validation for output formats and image sources.
|
|
1089
|
-
- Keep Python streaming work off the ASGI event loop and make Rust Anthropic SSE truly incremental so concurrent Claude Code requests are not stalled or buffered.
|
|
1090
|
-
- Preserve Rust streaming authentication, rate-limit, and overload errors after OAuth refresh instead of collapsing them to generic HTTP 500 failures.
|
|
1091
|
-
- Restrict the Python source distribution to release inputs so local agent state and unrelated runtime sources cannot enter PyPI artifacts.
|
|
1092
|
-
|
|
1093
|
-
### v0.6.0
|
|
1094
|
-
|
|
1095
|
-
- Add GPT-5.6 public alias plus Sol, Terra, and Luna capability metadata, model defaults, and Codex context-window behavior.
|
|
1096
|
-
- Refresh bundled context and default-effort metadata for current GPT-5.5, GPT-5.4, GPT-5.4 Mini, and GPT-5.2 catalog entries.
|
|
1097
|
-
- Update Responses Lite request bodies and headers across chat, compact, and inspection paths; Lite image generation is rejected explicitly because this proxy has no standalone image-tool executor.
|
|
1098
|
-
- Add `max`, Codex-compatible `ultra` to `max` wire conversion, future model-defined efforts, and `model_reasoning_effort` config support.
|
|
1099
|
-
- Add GPT-5.6 reasoning effort/context, `standard` mode compatibility, capability-gated image `original` detail, standard verbosity, cache accounting, and real backend `response_id` support across Python, TypeScript, and Rust.
|
|
1100
|
-
- Translate known `previous_response_id` values into bounded local full-history replay; reject Pro and non-null public cache policy/breakpoints or `safety_identifier` because the private Codex request contract has no equivalent field.
|
|
1101
|
-
- Reject hosted Multi-agent and Programmatic Tool Calling on the Chat facade until their native agent/program item lifecycle can be preserved.
|
|
1102
|
-
|
|
1103
|
-
### v0.5.2
|
|
1104
|
-
|
|
1105
|
-
- Support latest Codex root-level OAuth token files while keeping PAT-only, agent-identity-only, and Bedrock-only auth files explicitly unsupported.
|
|
1106
|
-
- Add shared model capability gating for Responses Lite, parallel tool calls, verbosity, and service-tier behavior across Python, TypeScript, and Rust.
|
|
1107
|
-
- Preserve encrypted reasoning state via top-level `reasoning.encrypted_content` include and add Codex metadata forwarding controls.
|
|
1108
|
-
|
|
1109
|
-
### v0.5.1
|
|
1110
|
-
|
|
1111
|
-
- Add official Codex CLI `originator` and versioned `User-Agent` headers for ChatGPT/Codex OAuth requests.
|
|
1112
|
-
- Resolve the latest `@openai/codex` version from npm at server startup, with `CODEX_AS_API_CODEX_CLI_VERSION` as an explicit override.
|
|
1113
|
-
|
|
1114
|
-
### v0.5.0
|
|
1115
|
-
|
|
1116
|
-
- Preserve Claude Code server-tool history (`server_tool_use`, `web_search_tool_result`, MCP/advisor-like result blocks) as backend context instead of dropping it on the next turn.
|
|
1117
|
-
- Support Anthropic `output_format` structured outputs by mapping JSON schema/object formats to OpenAI Responses `text.format`.
|
|
1118
|
-
- Preserve `document` and `search_result` content blocks inside tool results, keep Python streaming defaults aligned, and accept unsuffixed `web_search` server-tool types across Python, TypeScript, and Rust.
|
|
1119
|
-
|
|
1120
|
-
### v0.4.0
|
|
1121
|
-
|
|
1122
|
-
- Add Claude Code-compatible Anthropic hosted web search support by mapping `web_search_*` server tools to OpenAI Responses `web_search`.
|
|
1123
|
-
- Return `server_tool_use` and `web_search_tool_result` blocks so Claude Code can parse web search results reliably.
|
|
1124
|
-
- Prepare TypeScript package publishing to npmjs and GitHub Packages via GitHub Actions.
|
|
1125
|
-
|
|
1126
|
-
### v0.3.3
|
|
1127
|
-
|
|
1128
|
-
- Stop forwarding client `max_tokens` as Codex `max_output_tokens`, restoring Claude Code compatibility with the Codex OAuth backend.
|
|
1129
|
-
- Add Python, TypeScript, and Rust regression tests for the provider payload.
|
|
1130
|
-
|
|
1131
|
-
### v0.3.2
|
|
1132
|
-
|
|
1133
|
-
- Restore immediate Anthropic streaming so Claude Code receives events without waiting for the backend response to finish.
|
|
1134
|
-
- Use conservative local token estimates for `/v1/messages/count_tokens`; Codex OAuth has no count-only backend endpoint.
|
|
1135
|
-
- Keep real final streaming usage metadata in `message_delta`.
|
|
1136
|
-
|
|
1137
|
-
### v0.3.1
|
|
1138
|
-
|
|
1139
|
-
- Attempted real backend token counting for `/v1/messages/count_tokens` with `max_output_tokens: 0`; this is superseded by v0.3.2 because Codex OAuth rejects count-only requests.
|
|
1140
|
-
- Forward converted Anthropic tools, tool choice, stop sequences, and thinking/reasoning settings during token-count requests.
|
|
1141
|
-
- Propagate cumulative Anthropic streaming usage, including cache accounting, server tool use, and service tier metadata when available.
|
|
1142
|
-
- Pass `max_output_tokens` through provider requests across Python, TypeScript, and Rust.
|
|
1143
|
-
|
|
1144
|
-
### v0.3.0
|
|
1145
|
-
|
|
1146
|
-
- Read Codex CLI config from `CODEX_HOME` / `~/.codex/config.toml` across Python, TypeScript, and Rust.
|
|
1147
|
-
- Use the configured Codex backend model while preserving Anthropic client model names in `/v1/messages` responses.
|
|
1148
|
-
- Expose `context_window` and `auto_compact_token_limit` through `/health` and `/v1/messages/count_tokens`.
|
|
1149
|
-
- Add Anthropic-compatible `/v1/messages/count_tokens` and `/v1/messages/compact`.
|
|
1150
|
-
- Map context-window failures to Anthropic-style `400 invalid_request_error` responses and stream error events.
|
|
1151
|
-
|
|
1152
1006
|
## License
|
|
1153
1007
|
|
|
1154
1008
|
Apache License 2.0 — derived from [OpenAI Codex CLI](https://github.com/openai/codex) (Apache-2.0, Copyright 2025 OpenAI).
|
|
@@ -1,5 +1,41 @@
|
|
|
1
1
|
# Release Notes
|
|
2
2
|
|
|
3
|
+
## v0.10.2
|
|
4
|
+
|
|
5
|
+
### Realtime call & sideband deadlines and cross-runtime parity fixes
|
|
6
|
+
|
|
7
|
+
- Bound `createRealtimeCall` (`POST /v1/realtime/calls`) with a wall-clock deadline
|
|
8
|
+
(`REALTIME_CALL_TIMEOUT_MS = 15_000`, configurable via `timeoutMs` and `signal`)
|
|
9
|
+
covering the request and SDP answer read so silent or trickling upstream responses
|
|
10
|
+
abort with `504` instead of waiting indefinitely.
|
|
11
|
+
- Bound each WebSocket sideband join handshake in `relayRealtimeSideband` /
|
|
12
|
+
`attachRealtimeSideband` with a wall-clock deadline (`SIDEBAND_HANDSHAKE_TIMEOUT_MS = 10_000`,
|
|
13
|
+
configurable via `handshakeTimeoutMs`; cleared immediately on `open`), close the
|
|
14
|
+
client with `4504` (`realtime sideband unavailable (timeout)`) on timeout, and
|
|
15
|
+
reject the returned Promise on join or validation failure so callers never mark a
|
|
16
|
+
call live when the sideband WebSocket fails to attach. Forward the HTTP upgrade
|
|
17
|
+
`voice-session-id` header in `attachRealtimeSideband`.
|
|
18
|
+
- Reject whitespace-only `CODEX_AS_API_ADMIN_KEY` values with `503` (`management_disabled`)
|
|
19
|
+
in TypeScript and Rust, matching Python.
|
|
20
|
+
- Align Rust `POST /auth/login` request body limit to `4,096` bytes and return `400`
|
|
21
|
+
(`invalid_request`) when exceeded, matching Python and TypeScript.
|
|
22
|
+
- Derive session-scoped `prompt_cache_key` from `x-claude-code-session-id` on
|
|
23
|
+
`POST /v1/messages/compact` in Python and TypeScript, matching Rust.
|
|
24
|
+
|
|
25
|
+
## v0.10.1
|
|
26
|
+
|
|
27
|
+
### Preserve late system notes across Python, TypeScript and Rust chat histories
|
|
28
|
+
|
|
29
|
+
- Keep initial `system` messages in top-level `instructions`, and preserve `system`
|
|
30
|
+
messages appended after conversation history starts at their original positions as
|
|
31
|
+
`developer` input items across Python, TypeScript and Rust chat requests.
|
|
32
|
+
Appending per-turn runtime notes after tool results no longer mutates
|
|
33
|
+
`instructions` or invalidates the cached prefix of the prior conversation history.
|
|
34
|
+
- Remote compaction (`/v1/compact` and `/v1/messages/compact`) remains unchanged so
|
|
35
|
+
system directives continue to form `instructions` during compaction.
|
|
36
|
+
- Document current preview constraints of OpenAI's public "Sign in with ChatGPT"
|
|
37
|
+
(`chatgpt.tokens.use.direct`) route in `README.md`.
|
|
38
|
+
|
|
3
39
|
## v0.10.0
|
|
4
40
|
|
|
5
41
|
### Standalone browser login — no Codex installation required
|
|
@@ -4,13 +4,13 @@ build-backend = "hatchling.build"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "codex-as-api"
|
|
7
|
-
version = "0.10.
|
|
7
|
+
version = "0.10.2"
|
|
8
8
|
description = "Use ChatGPT/Codex OAuth as a local OpenAI-compatible API server."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
license = { text = "Apache-2.0" }
|
|
11
11
|
requires-python = ">=3.10"
|
|
12
12
|
authors = [
|
|
13
|
-
{ name = "
|
|
13
|
+
{ name = "CHOAM", email = "" },
|
|
14
14
|
]
|
|
15
15
|
keywords = ["openai", "codex", "chatgpt", "oauth", "api", "llm"]
|
|
16
16
|
classifiers = [
|
|
@@ -40,8 +40,8 @@ dev = ["pytest", "ruff", "mypy", "httpx", "types-regex", "tomli>=2.0.1"]
|
|
|
40
40
|
codex-as-api = "codex_as_api.server:main"
|
|
41
41
|
|
|
42
42
|
[project.urls]
|
|
43
|
-
Homepage = "https://github.com/
|
|
44
|
-
Repository = "https://github.com/
|
|
43
|
+
Homepage = "https://github.com/CHOAM-CO/codex-as-api"
|
|
44
|
+
Repository = "https://github.com/CHOAM-CO/codex-as-api"
|
|
45
45
|
|
|
46
46
|
[tool.hatch.build.targets.wheel]
|
|
47
47
|
packages = ["src/codex_as_api"]
|