codex-as-api 0.10.1__tar.gz → 1.0.0__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-0.10.1 → codex_as_api-1.0.0}/PKG-INFO +51 -203
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/README.md +47 -199
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/RELEASES.md +32 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/pyproject.toml +4 -4
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/scripts/check_package_versions.py +48 -2
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/scripts/check_release_artifacts.py +1 -1
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/__init__.py +1 -1
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/anthropic_adapter.py +397 -31
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/provider.py +689 -17
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/reasoning_signature.py +38 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/server.py +402 -21
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_anthropic_adapter.py +180 -106
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_package_versions.py +63 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_provider.py +314 -6
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_registry_release.py +16 -16
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_release_artifacts.py +3 -3
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_server.py +366 -22
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_standalone_login_api.py +4 -1
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/.gitignore +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/LICENSE +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/THIRD_PARTY_NOTICES.md +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/config/codex-upstream-contract.json +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/config/o200k_base.tiktoken +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/config/verified-hidden-models.json +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/requirements.txt +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/rust/THIRD_PARTY_NOTICES.md +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/scripts/check_no_static_model_catalog.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/scripts/check_registry_release.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/scripts/generate_rust_third_party_notices.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/scripts/package_rust_release.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/auth.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/browser_login.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/codex_config.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/codex_version.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/device_login.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/messages.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/model_capabilities.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/o200k_tokenizer.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/protocol.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/src/codex_as_api/strict_json.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/__init__.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/conftest.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/fixtures/o200k_base_encode_ordinary.json +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_auth.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_browser_login.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_codex_config.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_codex_version.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_device_login.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_login_commit_races.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_login_generation.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_login_http_real_store.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_management_body.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_messages.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_model_capabilities.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_no_static_model_catalog.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_o200k_tokenizer.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_protocol.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_rust_third_party_notices.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/tests/test_strict_json.py +0 -0
- {codex_as_api-0.10.1 → codex_as_api-1.0.0}/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.
|
|
3
|
+
Version: 1.0.0
|
|
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
|
|
@@ -35,16 +35,16 @@ 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
|
|
|
45
45
|
## Features
|
|
46
46
|
|
|
47
|
-
- **OpenAI & Anthropic compatible** — Chat Completions, Messages, and `GET /v1/models`
|
|
47
|
+
- **OpenAI & Anthropic compatible** — Responses (`/v1/responses`, `/v1/responses/compact`), Chat Completions, Messages, and `GET /v1/models`
|
|
48
48
|
- **Claude Code ready** — use Codex models directly from Claude Code CLI
|
|
49
49
|
- **Streaming** — full SSE streaming for both OpenAI and Anthropic protocols
|
|
50
50
|
- **Tool calling** — function calls, tool results, and parallel tool calls
|
|
@@ -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
|
|
@@ -298,7 +298,7 @@ These three upstream-hidden models returned real responses on October 24, 2026,
|
|
|
298
298
|
|
|
299
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.
|
|
300
300
|
|
|
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/
|
|
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.
|
|
302
302
|
|
|
303
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.
|
|
304
304
|
|
|
@@ -392,7 +392,7 @@ it; continuation uses the complete provider Response items saved behind
|
|
|
392
392
|
|
|
393
393
|
Anthropic Messages compatibility endpoint for any client, including Claude Code and Scheherazade. It supports streaming (`stream: true`) and non-streaming without a client-identification header. A live Codex model ID routes directly. A `claude-*` facade ID uses the configured Codex backend model, or the authenticated catalog default when no backend model is configured. Other unknown IDs fail. The response preserves the client-supplied model name.
|
|
394
394
|
|
|
395
|
-
`max_tokens` is a required positive compatibility hint: the private Codex OAuth transport does not expose a matching output-token cap. Valid ephemeral `cache_control` hints are accepted and stripped for every client; their TTL and breakpoint semantics are not forwarded. An optional `x-claude-code-session-id` provides stable hashed cache affinity, not permission to use the endpoint. `metadata
|
|
395
|
+
`max_tokens` is a required positive compatibility hint: the private Codex OAuth transport does not expose a matching output-token cap. Valid ephemeral `cache_control` hints are accepted and stripped for every client; their TTL and breakpoint semantics are not forwarded. An optional `x-claude-code-session-id` provides stable hashed cache affinity, not permission to use the endpoint. Anthropic `metadata` objects with string or null entries (including `user_id` and additional client metadata keys) are accepted and not forwarded.
|
|
396
396
|
|
|
397
397
|
```bash
|
|
398
398
|
curl http://localhost:18080/v1/messages \
|
|
@@ -435,12 +435,21 @@ On the following assistant-history request, recognized envelopes restore the ori
|
|
|
435
435
|
|
|
436
436
|
Multiple summary parts share one thinking block per reasoning item. Encrypted-only items use an empty thinking string with a non-empty signature. Streaming sends `signature_delta` before the matching `content_block_stop`. `display: "omitted"` hides the summary but can preserve an encrypted-only signed block for continuity; disabled thinking suppresses thinking output. Items without encrypted content retain unsigned-summary compatibility behavior.
|
|
437
437
|
|
|
438
|
-
Hosted Anthropic `web_search` declarations
|
|
439
|
-
|
|
440
|
-
|
|
441
|
-
|
|
442
|
-
|
|
443
|
-
|
|
438
|
+
Hosted Anthropic `web_search` declarations (`web_search`, `web_search_20250305`,
|
|
439
|
+
and `web_search_20260209`, including `allowed_domains`, `user_location`,
|
|
440
|
+
`tool_choice: {"type": "tool", "name": "web_search"}`, and positive integer
|
|
441
|
+
`max_uses`, which is validated and accepted for Anthropic compatibility without
|
|
442
|
+
upstream enforcement) are supported when Classic Responses is active
|
|
443
|
+
(`responses_lite: false`, or `"auto"` on models that use Classic Responses) and
|
|
444
|
+
map to Codex hosted `web_search` with `web_search_call.action.sources`,
|
|
445
|
+
projecting `search`, `open_page`, and `find_in_page` `web_search_call` actions into
|
|
446
|
+
`server_tool_use` and `web_search_tool_result` blocks alongside
|
|
447
|
+
`usage.server_tool_use` (and preserving the raw `web_search_call` items on `/v1/responses`).
|
|
448
|
+
`blocked_domains` and `strict: true` on `web_search`,
|
|
449
|
+
or hosted `web_search` when Responses Lite is active, return HTTP 400. Replayed
|
|
450
|
+
`server_tool_use`, `web_search_tool_result`, text/content `document`, and
|
|
451
|
+
`search_result` blocks in conversation history are rendered into structured
|
|
452
|
+
text context.
|
|
444
453
|
|
|
445
454
|
Completed responses require authoritative final token usage and a real
|
|
446
455
|
`response.completed` event. Emitted function calls map to `tool_calls` or
|
|
@@ -512,7 +521,7 @@ curl http://localhost:18080/v1/messages/count_tokens \
|
|
|
512
521
|
|
|
513
522
|
### `POST /v1/messages/compact`
|
|
514
523
|
|
|
515
|
-
Anthropic-compatible alias for remote conversation compaction. Accepts Anthropic Messages-shaped bodies and returns compacted checkpoint content. Live GPT model IDs select the matching backend model, and Claude Code effort and Fast Mode controls use the same mappings as `/v1/messages`. Its
|
|
524
|
+
Anthropic-compatible alias for remote conversation compaction. Accepts Anthropic Messages-shaped bodies and returns compacted checkpoint content. Live GPT model IDs select the matching backend model, and Claude Code effort and Fast Mode controls use the same mappings as `/v1/messages`. Its optional non-forwarded `max_tokens` (validated as a positive integer when provided), sampling/stop compatibility fields (`temperature`, `top_p`, `top_k`, empty `stop_sequences: []`), `tool_choice`, and valid `cache_control` hints are accepted for every client; the optional Claude Code session header only supplies cache affinity.
|
|
516
525
|
|
|
517
526
|
### `POST /v1/images/generations`
|
|
518
527
|
|
|
@@ -661,7 +670,7 @@ Requests can instead use the Responses-shaped effort/context object:
|
|
|
661
670
|
- Responses Lite uses `all_turns` as the Codex wire default. An explicitly different context is rejected instead of silently overwritten; use `responses_lite: false` when the backend route supports classic Responses and another context is required.
|
|
662
671
|
- Remote compact keeps its existing private Codex `reasoning_effort` field but does not accept public `reasoning.mode` or `reasoning.context`.
|
|
663
672
|
|
|
664
|
-
Anthropic `thinking` values map as `enabled → high` and `adaptive → medium`. Disabled thinking selects `none` when available, otherwise the lowest supported catalog effort, and suppresses reasoning output; it does not claim to eliminate a model's internal reasoning. `output_config.effort` takes precedence over adaptive or enabled thinking and supports `low`, `medium`, `high`, `xhigh`, and `max`. Call-level disabled thinking takes precedence over ambient `output_config.effort` for auxiliary calls such as WebFetch. Explicit unsupported effort requests still fail. Hosted Anthropic
|
|
673
|
+
Anthropic `thinking` values map as `enabled → high` and `adaptive → medium`. Disabled thinking selects `none` when available, otherwise the lowest supported catalog effort, and suppresses reasoning output; it does not claim to eliminate a model's internal reasoning. `output_config.effort` takes precedence over adaptive or enabled thinking and supports `low`, `medium`, `high`, `xhigh`, and `max`. Call-level disabled thinking takes precedence over ambient `output_config.effort` for auxiliary calls such as WebFetch. Explicit unsupported effort requests still fail. Hosted Anthropic `web_search` declarations are bridged when Classic Responses is active (`responses_lite: false`, or `"auto"` on Classic-default models), while Responses Lite rejects hosted `web_search` with HTTP 400. The supported reasoning context and verbosity extensions are also available on `/v1/messages`, image generation, and inspection requests; Pro, non-null public cache policy/breakpoints, and any supplied `safety_identifier` fail explicitly on the private Codex provider.
|
|
665
674
|
|
|
666
675
|
Anthropic's `JSONOutputFormat` carries a schema but no name. When Claude Code
|
|
667
676
|
sends that official shape, the proxy follows pinned Codex 0.153.3 and supplies
|
|
@@ -670,10 +679,11 @@ or fallback role. An explicit proxy-extension name is preserved only when it is
|
|
|
670
679
|
already a valid Responses format name, and is never sanitized or truncated.
|
|
671
680
|
|
|
672
681
|
The pinned official Codex HTTP request has no `stop` field. Omitted OpenAI
|
|
673
|
-
`stop`,
|
|
674
|
-
any
|
|
675
|
-
|
|
676
|
-
|
|
682
|
+
`stop`, explicit `stop: null`, and empty `stop: []` are omitted from the private
|
|
683
|
+
request as no-ops; any non-empty OpenAI `stop` value returns HTTP 400. Anthropic
|
|
684
|
+
`stop_sequences` allows omission and empty `stop_sequences: []`, while explicit
|
|
685
|
+
`stop_sequences: null` or any non-empty array returns HTTP 400 because the
|
|
686
|
+
private transport cannot apply stop sequences.
|
|
677
687
|
|
|
678
688
|
The mapping is intentionally transport-aware:
|
|
679
689
|
|
|
@@ -683,7 +693,7 @@ The mapping is intentionally transport-aware:
|
|
|
683
693
|
| `prompt_cache_key` | Forward an explicit value; otherwise use non-empty `client_metadata.session_id` |
|
|
684
694
|
| Non-null `prompt_cache_options` / breakpoint | HTTP 400; `null` is omitted and there is no private field |
|
|
685
695
|
| Any supplied `safety_identifier` | HTTP 400; the optional field is non-nullable and OAuth account/thread IDs are not semantic aliases |
|
|
686
|
-
| Non-
|
|
696
|
+
| Non-empty `stop` / `stop_sequences` (or `stop_sequences: null`) | HTTP 400; empty `[]` (and OpenAI `stop: null`) are omitted |
|
|
687
697
|
| `previous_response_id` | Resolve locally and replay complete Response history over HTTP |
|
|
688
698
|
| `service_tier: "fast"` | Send `service_tier: "priority"` |
|
|
689
699
|
| `service_tier: "default"` | Omit the field |
|
|
@@ -894,12 +904,15 @@ CODEX_AS_API_MODEL=gpt-5.6-terra \
|
|
|
894
904
|
codex-as-api
|
|
895
905
|
```
|
|
896
906
|
|
|
897
|
-
Claude Code hosted
|
|
898
|
-
|
|
899
|
-
|
|
900
|
-
|
|
901
|
-
|
|
902
|
-
|
|
907
|
+
Claude Code hosted `web_search` (`web_search`, `web_search_20250305`, and
|
|
908
|
+
`web_search_20260209`) is supported when Classic Responses is active
|
|
909
|
+
(`responses_lite: false`, or `"auto"` on models whose catalog sets Classic
|
|
910
|
+
Responses), mapping Codex `web_search` calls and `web_search_call.action.sources`
|
|
911
|
+
into Anthropic `server_tool_use` and `web_search_tool_result` blocks alongside
|
|
912
|
+
`usage.server_tool_use`. On Responses Lite models under `responses_lite: "auto"`,
|
|
913
|
+
hosted `web_search` returns HTTP 400 unless `responses_lite: false` is valid
|
|
914
|
+
and supplied. WebFetch remains a client-side auxiliary request and retains the
|
|
915
|
+
disabled-thinking precedence behavior when process-level effort is enabled.
|
|
903
916
|
|
|
904
917
|
To keep the built-in Fable, Opus, Sonnet, and Haiku rows and append one GPT row, launch the GPT-routed Claude Code process with these variables:
|
|
905
918
|
|
|
@@ -981,10 +994,12 @@ The provider handles:
|
|
|
981
994
|
## Release & package publishing
|
|
982
995
|
|
|
983
996
|
The `Release` GitHub Actions workflow runs on a pushed `v*` tag. It rejects a
|
|
984
|
-
tag that does not exactly match every Python, npm, and Rust version surface
|
|
997
|
+
tag that does not exactly match every Python, npm, and Rust version surface or
|
|
998
|
+
lacks a matching `## v<version>` section in [RELEASES.md](RELEASES.md),
|
|
985
999
|
runs the complete cross-runtime gate, and then publishes PyPI, npmjs, the
|
|
986
1000
|
scoped GitHub npm package, deterministic platform Rust ZIP archives, and one
|
|
987
|
-
GitHub Release
|
|
1001
|
+
GitHub Release whose body is populated from that `RELEASES.md` section. Each
|
|
1002
|
+
Rust archive contains the platform binary, `LICENSE`, and
|
|
988
1003
|
the generated `THIRD_PARTY_NOTICES.md` covering both bundled tokenizer data and
|
|
989
1004
|
the locked normal dependency graph across supported release targets.
|
|
990
1005
|
|
|
@@ -994,10 +1009,10 @@ registry setup is still required:
|
|
|
994
1009
|
|
|
995
1010
|
- Create GitHub environments named `pypi` and `npm`; add deployment protection
|
|
996
1011
|
rules if desired.
|
|
997
|
-
- On PyPI, add a trusted publisher for project `codex-as-api`, owner `
|
|
1012
|
+
- On PyPI, add a trusted publisher for project `codex-as-api`, owner `CHOAM-CO`,
|
|
998
1013
|
repository `codex-as-api`, workflow `release.yml`, environment `pypi`.
|
|
999
1014
|
- On npmjs, add a GitHub Actions trusted publisher for package `codex-as-api`,
|
|
1000
|
-
owner `
|
|
1015
|
+
owner `CHOAM-CO`, repository `codex-as-api`, workflow `release.yml`,
|
|
1001
1016
|
environment `npm`, with `npm publish` allowed.
|
|
1002
1017
|
- GitHub Packages and GitHub Releases use the job-scoped automatic
|
|
1003
1018
|
`GITHUB_TOKEN`; repository policy must allow the workflow's declared
|
|
@@ -1006,9 +1021,9 @@ registry setup is still required:
|
|
|
1006
1021
|
After bumping every version and passing the local gate, push the matching tag:
|
|
1007
1022
|
|
|
1008
1023
|
```bash
|
|
1009
|
-
python scripts/check_package_versions.py --tag
|
|
1010
|
-
git tag
|
|
1011
|
-
git push origin
|
|
1024
|
+
python scripts/check_package_versions.py --tag v1.0.0
|
|
1025
|
+
git tag v1.0.0
|
|
1026
|
+
git push origin v1.0.0
|
|
1012
1027
|
```
|
|
1013
1028
|
|
|
1014
1029
|
## Tests
|
|
@@ -1036,173 +1051,6 @@ npm install
|
|
|
1036
1051
|
npm test
|
|
1037
1052
|
```
|
|
1038
1053
|
|
|
1039
|
-
|
|
1040
|
-
## Release Notes
|
|
1041
|
-
|
|
1042
|
-
### v0.10.1
|
|
1043
|
-
|
|
1044
|
-
- Preserve mid-history `system` messages added after conversation history starts as in-place `developer` input items across Python, TypeScript, and Rust chat requests, keeping the earlier conversation history prefix stable for prompt caching while leaving remote compaction (`compact_messages`) unchanged.
|
|
1045
|
-
- Document current preview limitations of OpenAI's public "Sign in with ChatGPT" (`chatgpt.tokens.use.direct`) route and why the Codex OAuth transport remains the default.
|
|
1046
|
-
|
|
1047
|
-
### v0.10.0
|
|
1048
|
-
|
|
1049
|
-
- Sign in directly with ordinary browser OAuth; no Codex installation or existing
|
|
1050
|
-
Codex credentials are required. Device-code login is an explicit remote option.
|
|
1051
|
-
- Add independent admin-key-protected login start/status/cancel endpoints and a
|
|
1052
|
-
private application-owned credential store across Python, TypeScript and Rust.
|
|
1053
|
-
- Upgrading requires a new codex-as-api login; existing Codex credentials remain
|
|
1054
|
-
untouched. See [Standalone login](#standalone-login) and [full release notes](RELEASES.md).
|
|
1055
|
-
|
|
1056
|
-
### v0.8.0
|
|
1057
|
-
|
|
1058
|
-
- Add ChatGPT Live realtime voice to the TypeScript runtime: `POST
|
|
1059
|
-
/v1/realtime/calls` creates an OAuth-authorized call from a client-owned WebRTC
|
|
1060
|
-
SDP offer and returns the SDP answer, and `WS /v1/realtime/sideband/{call_id}`
|
|
1061
|
-
relays the OpenAI realtime event sideband for headless clients. Audio stays on
|
|
1062
|
-
the client's media plane.
|
|
1063
|
-
- Mirror the pinned Codex call contract exactly: one identity in
|
|
1064
|
-
`x-session-id`/`session-id`/`thread-id`, the `openai-alpha: quicksilver=v2`
|
|
1065
|
-
handshake, `intent=quicksilver&architecture=avas`, and a session payload that
|
|
1066
|
-
never chooses its own voice-session id.
|
|
1067
|
-
- Document realtime as a TypeScript-only capability; Python and Rust remain
|
|
1068
|
-
HTTP-only in this release instead of claiming cross-runtime parity.
|
|
1069
|
-
|
|
1070
|
-
### v0.7.0
|
|
1071
|
-
|
|
1072
|
-
- Replace bundled model capability tables with a fresh authenticated Codex
|
|
1073
|
-
catalog and add `GET /v1/models` across all three runtimes.
|
|
1074
|
-
- Remove stale/unknown-model fallbacks, fabricated Anthropic signatures, silent
|
|
1075
|
-
request downgrades, and malformed upstream recovery; return typed failures.
|
|
1076
|
-
- Reject hosted Anthropic WebSearch instead of translating incomplete OpenAI
|
|
1077
|
-
search lifecycle data, and require authoritative completion and final usage.
|
|
1078
|
-
- Keep immediate Claude Code `message_start` delivery without inventing token
|
|
1079
|
-
counts; final Anthropic fields are emitted only from authoritative provider data.
|
|
1080
|
-
- Scope continuation history and catalog state to the authenticated account,
|
|
1081
|
-
validate before streaming, and make `/health` report actual readiness.
|
|
1082
|
-
- Require the official nested `tokens` auth layout, remove root-level and
|
|
1083
|
-
noncanonical auth aliases, and reject external-host `chatgptAuthTokens`.
|
|
1084
|
-
- Pin the validated private contract to Codex `0.153.3` and publish the exact
|
|
1085
|
-
tested artifacts to PyPI, npmjs, GitHub Packages, and GitHub Releases.
|
|
1086
|
-
- Install Python server dependencies with the base package so the default
|
|
1087
|
-
`pip install codex-as-api` command produces a runnable server.
|
|
1088
|
-
|
|
1089
|
-
### v0.6.5
|
|
1090
|
-
|
|
1091
|
-
- Align GPT-5.6 alias, Sol, Terra, and Luna context limits with Codex `0.147.0`
|
|
1092
|
-
at 272,000 tokens and derive the 244,800-token automatic compact threshold.
|
|
1093
|
-
- Pin the verified upstream request, Responses Lite, header, and model contract
|
|
1094
|
-
to `openai/codex` commit
|
|
1095
|
-
`be6e8eac029b183056b7e4402879f15d2c85f61b`; track its new fractional SSE
|
|
1096
|
-
rollout-budget field without claiming an unapproved facade mapping.
|
|
1097
|
-
- Refresh OAuth credentials five minutes before expiry, coalesce concurrent
|
|
1098
|
-
refreshes, reload changed credentials after 401, and preserve upstream HTTP
|
|
1099
|
-
status codes across all three servers.
|
|
1100
|
-
- Replace startup npm version discovery with deterministic compatibility and
|
|
1101
|
-
package-version identity in the Codex-style `User-Agent`.
|
|
1102
|
-
- Add full cross-runtime CI plus tag-gated OIDC publishing to PyPI and npmjs,
|
|
1103
|
-
`GITHUB_TOKEN` publishing to GitHub Packages, and GitHub Release assets.
|
|
1104
|
-
|
|
1105
|
-
### v0.6.4
|
|
1106
|
-
|
|
1107
|
-
- Remove process-global generated Codex session/thread identities and require
|
|
1108
|
-
explicit session identity when Codex metadata mode is enabled.
|
|
1109
|
-
- Resolve Chat cache affinity as explicit `prompt_cache_key`, then
|
|
1110
|
-
`client_metadata.session_id`, then omission.
|
|
1111
|
-
- Derive Claude Code cache affinity from `x-claude-code-session-id` without
|
|
1112
|
-
fabricating Codex metadata; keep normal Anthropic Messages stateless.
|
|
1113
|
-
- Validate and strip Anthropic `cache_control` compatibility hints without
|
|
1114
|
-
claiming unavailable breakpoint or TTL semantics.
|
|
1115
|
-
- Update current compatibility evidence to Codex `0.145.0`, Codex `main` at
|
|
1116
|
-
`bd2de422aa287b97b06ca6425a10935bcf1b3731`, and Claude Code `2.1.220`.
|
|
1117
|
-
|
|
1118
|
-
### v0.6.3
|
|
1119
|
-
|
|
1120
|
-
- Accept Claude Code auxiliary requests that combine ambient `output_config.effort` with call-level `thinking.disabled`.
|
|
1121
|
-
- Give explicit disabled thinking precedence and send Codex reasoning effort `none` instead of returning HTTP 400.
|
|
1122
|
-
- Document and test `CODEX_AS_API_RESPONSES_LITE=off` for Claude Code hosted WebSearch on GPT-5.6; WebFetch needs no Responses Lite override.
|
|
1123
|
-
- Preserve fail-loudly validation for invalid effort values and unsupported `output_config` fields.
|
|
1124
|
-
- Add Python, TypeScript, and Rust adapter regressions, streamed `/v1/messages` coverage, and a `/count_tokens` regression for the WebSearch/WebFetch request shape.
|
|
1125
|
-
|
|
1126
|
-
### v0.6.2
|
|
1127
|
-
|
|
1128
|
-
- Fix `/v1/messages/count_tokens` overcounting that could trigger Claude Code autocompaction on every turn.
|
|
1129
|
-
- Count normalized model-visible messages, tool calls, reasoning, tool-result metadata, and tool schemas once; remove the second full raw-request-body addition.
|
|
1130
|
-
- Port official `tiktoken` `o200k_base` ordinary encoding into Python, TypeScript, and Rust without adding a `tiktoken` package dependency or runtime rank download.
|
|
1131
|
-
- Match the upstream Unicode text split and BPE merge ranks last checked on 2026-07-14 against `tiktoken` 0.13.0 at `08a5f3b`.
|
|
1132
|
-
- Keep image input cost separate and count inline base64 images once rather than adding both a fixed image estimate and the base64 request bytes.
|
|
1133
|
-
- 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.
|
|
1134
|
-
- Validate the release with Claude Code `2.1.209` and real Codex OAuth chat.
|
|
1135
|
-
|
|
1136
|
-
### v0.6.1
|
|
1137
|
-
|
|
1138
|
-
- Re-audit official Codex `0.144.4` and `main` at `393f64565ab46f09d99ca4d9bd973537e72a114b`, plus Claude Code `2.1.208`, before publishing.
|
|
1139
|
-
- 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.
|
|
1140
|
-
- 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.
|
|
1141
|
-
- 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.
|
|
1142
|
-
- Keep Messages, token counting, and remote compaction aligned on effective model limits and fail-loudly validation for output formats and image sources.
|
|
1143
|
-
- 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.
|
|
1144
|
-
- Preserve Rust streaming authentication, rate-limit, and overload errors after OAuth refresh instead of collapsing them to generic HTTP 500 failures.
|
|
1145
|
-
- Restrict the Python source distribution to release inputs so local agent state and unrelated runtime sources cannot enter PyPI artifacts.
|
|
1146
|
-
|
|
1147
|
-
### v0.6.0
|
|
1148
|
-
|
|
1149
|
-
- Add GPT-5.6 public alias plus Sol, Terra, and Luna capability metadata, model defaults, and Codex context-window behavior.
|
|
1150
|
-
- 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.
|
|
1151
|
-
- 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.
|
|
1152
|
-
- Add `max`, Codex-compatible `ultra` to `max` wire conversion, future model-defined efforts, and `model_reasoning_effort` config support.
|
|
1153
|
-
- 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.
|
|
1154
|
-
- 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.
|
|
1155
|
-
- Reject hosted Multi-agent and Programmatic Tool Calling on the Chat facade until their native agent/program item lifecycle can be preserved.
|
|
1156
|
-
|
|
1157
|
-
### v0.5.2
|
|
1158
|
-
|
|
1159
|
-
- Support latest Codex root-level OAuth token files while keeping PAT-only, agent-identity-only, and Bedrock-only auth files explicitly unsupported.
|
|
1160
|
-
- Add shared model capability gating for Responses Lite, parallel tool calls, verbosity, and service-tier behavior across Python, TypeScript, and Rust.
|
|
1161
|
-
- Preserve encrypted reasoning state via top-level `reasoning.encrypted_content` include and add Codex metadata forwarding controls.
|
|
1162
|
-
|
|
1163
|
-
### v0.5.1
|
|
1164
|
-
|
|
1165
|
-
- Add official Codex CLI `originator` and versioned `User-Agent` headers for ChatGPT/Codex OAuth requests.
|
|
1166
|
-
- Resolve the latest `@openai/codex` version from npm at server startup, with `CODEX_AS_API_CODEX_CLI_VERSION` as an explicit override.
|
|
1167
|
-
|
|
1168
|
-
### v0.5.0
|
|
1169
|
-
|
|
1170
|
-
- 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.
|
|
1171
|
-
- Support Anthropic `output_format` structured outputs by mapping JSON schema/object formats to OpenAI Responses `text.format`.
|
|
1172
|
-
- 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.
|
|
1173
|
-
|
|
1174
|
-
### v0.4.0
|
|
1175
|
-
|
|
1176
|
-
- Add Claude Code-compatible Anthropic hosted web search support by mapping `web_search_*` server tools to OpenAI Responses `web_search`.
|
|
1177
|
-
- Return `server_tool_use` and `web_search_tool_result` blocks so Claude Code can parse web search results reliably.
|
|
1178
|
-
- Prepare TypeScript package publishing to npmjs and GitHub Packages via GitHub Actions.
|
|
1179
|
-
|
|
1180
|
-
### v0.3.3
|
|
1181
|
-
|
|
1182
|
-
- Stop forwarding client `max_tokens` as Codex `max_output_tokens`, restoring Claude Code compatibility with the Codex OAuth backend.
|
|
1183
|
-
- Add Python, TypeScript, and Rust regression tests for the provider payload.
|
|
1184
|
-
|
|
1185
|
-
### v0.3.2
|
|
1186
|
-
|
|
1187
|
-
- Restore immediate Anthropic streaming so Claude Code receives events without waiting for the backend response to finish.
|
|
1188
|
-
- Use conservative local token estimates for `/v1/messages/count_tokens`; Codex OAuth has no count-only backend endpoint.
|
|
1189
|
-
- Keep real final streaming usage metadata in `message_delta`.
|
|
1190
|
-
|
|
1191
|
-
### v0.3.1
|
|
1192
|
-
|
|
1193
|
-
- 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.
|
|
1194
|
-
- Forward converted Anthropic tools, tool choice, stop sequences, and thinking/reasoning settings during token-count requests.
|
|
1195
|
-
- Propagate cumulative Anthropic streaming usage, including cache accounting, server tool use, and service tier metadata when available.
|
|
1196
|
-
- Pass `max_output_tokens` through provider requests across Python, TypeScript, and Rust.
|
|
1197
|
-
|
|
1198
|
-
### v0.3.0
|
|
1199
|
-
|
|
1200
|
-
- Read Codex CLI config from `CODEX_HOME` / `~/.codex/config.toml` across Python, TypeScript, and Rust.
|
|
1201
|
-
- Use the configured Codex backend model while preserving Anthropic client model names in `/v1/messages` responses.
|
|
1202
|
-
- Expose `context_window` and `auto_compact_token_limit` through `/health` and `/v1/messages/count_tokens`.
|
|
1203
|
-
- Add Anthropic-compatible `/v1/messages/count_tokens` and `/v1/messages/compact`.
|
|
1204
|
-
- Map context-window failures to Anthropic-style `400 invalid_request_error` responses and stream error events.
|
|
1205
|
-
|
|
1206
1054
|
## License
|
|
1207
1055
|
|
|
1208
1056
|
Apache License 2.0 — derived from [OpenAI Codex CLI](https://github.com/openai/codex) (Apache-2.0, Copyright 2025 OpenAI).
|