Skip to content

feat(python): provider media for Gemini, Vertex, Bedrock, LiteLLM and compatible chat - #697

Draft
nik13 wants to merge 15 commits into
feat/attachments-python-openai-anthropicfrom
feat/attachments-python-providers
Draft

nik13 wants to merge 15 commits into
feat/attachments-python-openai-anthropicfrom
feat/attachments-python-providers

Conversation

@nik13

@nik13 nik13 commented Nov 12, 2026 •

Copy link
Copy Markdown
Contributor

Part of TH-2379 (Linear).

Summary

Python provider media for Gemini, Vertex, Bedrock, LiteLLM and the OpenAI-compatible chat packages (work packages M05, M06, M08, M10 and M12). It mirrors the TypeScript @traceai/provider-media helper (#683): one shared helper, thin per-package bindings, and a capability table that is also the support matrix.

Stacked on #690 (feat/attachments-python-openai-anthropic). It reuses that PR's media_capture helper; it does not fork it.

  • New fi_instrumentation.attachments.provider_media (adapter-facing, not in attachments.__all__):
    • ProviderMedia(manager, span, namespace, trace_config=) has typed walkers: gemini_contents / gemini_response, compatible_messages(messages, dialect), ollama_messages / ollama_images, generated_images, bedrock_converse, anthropic_messages, bedrock_invoke_request / bedrock_invoke_response and imagen_predictions.
    • Helpers: manager_from_kwargs, claim(context), sniff, consistent and capabilities(package).
    • PROVIDER_MEDIA_CAPABILITIES is the support matrix.
  • Shared context keys: the first commit applies the media_capture.py hunks from feat/attachments-python-auto (1b3072b + 0b2eb4b) unchanged, byte for byte. That brings CAPTURE_POLICY_KEY, OWNER_KEY, owned_elsewhere and merge_capture_policies, so merging that branch later changes nothing here. Every walker goes through media_capture.local_capture_policy and Observation. As a result the per-call policy key and the owner key apply to provider media, and no binding writes either key as a string literal, except the deny-all fallback in traceai_xai / traceai_minimax, used only if provider_media.deferred cannot be loaded.
  • Bindings: each instrumentor accepts attachments= (and auto's attachment_manager=). Each package exports <name>_attachment_capabilities lazily.

Support matrix

Package Captured (typed sources only) Not captured (fail closed)
traceai_google_genai models.generate_content: input inline_data Parts (image, audio, video, PDF) and output candidate inline_data (generated images, audio, files). generate_content_stream: input only file_data / gs:// URIs, provider files, PIL images, streamed output, generate_images, generate_videos, Live, audio/L16 PCM
traceai_vertexai Sync gapic PredictionServiceClient only: generate_content (v1, v1beta1) input and output inline_data, and Imagen predict output (bytesBase64Encoded with a declared image type, only on /models/imagen-* / imagegeneration endpoints) file_data / gs://, streamed output, the async client, any other prediction endpoint. GenerativeModel and ImageGenerationModel run through the same client but have no fixture yet, so the descriptor does not claim them
traceai_bedrock Converse image / document byte blocks for Claude 3/4 and Nova lite/pro/premier. One toolResult level: a computer-use screenshot is recognized only when a typed computer_YYYYMMDD tool is declared AND an assistant toolUse of that tool has the ID. Nova video bytes. InvokeModel: Anthropic Messages bodies, Nova bodies, Titan Image Generator / Nova Canvas input images and output images, read from the body the instrumentor already buffered S3 locations, ConverseStream, InvokeModelWithResponseStream, unknown model bodies (never scraped), Claude video
traceai_litellm completion / acompletion / completion_with_retries chat image_url data URIs, input_audio, and PDF file (file.file_data) / input_file (file_data) data URIs. image_generation / aimage_generation b64_json output URL results (never fetched), provider file IDs, application/octet-stream files, streamed output, image edit, speech, transcription
traceai_groq (chat) chat.completions.create image_url data URIs (image only, as in TypeScript) chat input_audio, remote URLs, output media. Speech and transcription stay as in #690
traceai_fireworks chat image_url data URIs. It defers the whole call to traceai_openai while that is enabled (pre-existing rule) remote URLs, audio, output media
traceai_xai chat image_url data URIs. With attachments enabled for it and traceai_openai enabled with an active attachment manager, it hands the whole call to traceai_openai under its own local deny; with attachments off it traces exactly as before remote URLs, audio, output media
traceai_vllm (legacy VLLMInstrumentor) chat image_url data URIs and input_audio (wav/mp3). It already defers to traceai_openai while that is enabled remote URLs, video_url, output media
traceai_minimax chat image_url data URIs (MiniMax-M3 / M3.1-Flash accept images; there is no audio input). With attachments enabled for it and traceai_openai enabled with an active attachment manager, it hands the whole call to traceai_openai under its own local deny; with attachments off it traces exactly as before remote URLs, video_url, audio, output media, native speech/image/video (X04)
traceai_mistralai chat.complete / stream (and async) image_url (string or object) data URIs, in both the 1.x wrapper path and the 3.6+ native-hook path OCR, document_url, input_audio, remote URLs
traceai_openrouter chat.send image_url data URIs and input_audio (wav/mp3) remote URLs, output media
traceai_portkey chat image_url data URIs and input_audio the generation API, remote URLs
traceai_ollama chat messages[].images and generate images: bytes or base64, typed by magic bytes image paths (never opened), streamed output
traceai_together nothing (descriptor has no operations) traceai_together installs no chat wrapper today: its install check looks for chat on the Together class, where it is an instance attribute. The binding is wired, but no media is claimed until that pre-existing bug is fixed
traceai_deepseek, traceai_cerebras nothing (no qualified media surface) tracing only

Privacy defaults (unchanged)

  • Off unless enabled. Without a manager, or with a disabled or pre-attachments core, no attachment module is imported and tracing is unchanged. A subprocess probe instruments every package and checks this.
  • Admission before any read. manager.preflight runs before a source is read, decoded or sniffed.
    • A data-URI gate denied only for size is re-admitted once at the exact size.
    • A sibling exact type is re-admitted under that type.
    • Unused tickets are released.
  • The declared type must match the bytes. After admission, the leading bytes must match the declared type. A PNG labelled audio/wav is not uploaded, so a wrong label cannot get past hide_input_images.
  • Deny wins. Local TraceConfig hide/PII flags, the per-call context policy and the owner mark all apply.
  • No fetching. No URL, gs://, S3, provider file or filesystem path is ever fetched or opened.
  • Owner key. LiteLLM runs its downstream SDK call under media_capture.OWNER_KEY, so a nested OpenAI instrumentor uploads nothing more. LiteLLM therefore captures every chat part kind the OpenAI instrumentor would have captured there (images, PDF files) plus input_audio. No other binding claims. Every observing call owns its own occurrence; there is no cross-call claim.
  • One owner per OpenAI-SDK call. traceai_xai, traceai_minimax, traceai_fireworks and traceai_vllm wrap the same openai method as traceai_openai. traceai_fireworks and traceai_vllm defer the whole call to it while it is enabled (their existing rule). traceai_xai and traceai_minimax defer the whole call only while attachments are enabled for them AND traceai_openai holds an active attachment manager (OpenAIInstrumentor.captures_attachments()); with attachments off they keep both spans, exactly as on base, and next to an OpenAI instrumentor without a manager they trace and capture the call themselves. Either setup order works, and each part is uploaded once.
  • Deny propagation. A deferred xAI/MiniMax call runs under provider_media.deferred(config): the binding's local TraceConfig becomes a restriction-only CapturePolicy (pii_redaction denies all; an input hide flag or hide_input_images denies inputs; an output hide flag denies outputs), merged deny-wins with any per-call policy under media_capture.CAPTURE_POLICY_KEY. traceai_openai therefore cannot capture more than the binding allows, whatever its own TraceConfig. A CapturePolicy cannot single out images, so hide_input_images on the binding also denies input files on the deferred call.
  • Type checks. video/mp4 needs an MP4 ISO-BMFF major brand (isom, iso2–iso6, mp41, mp42, avc1, dash, M4V , mmp4). A HEIC/AVIF still image, QuickTime or M4A labelled video/mp4 is refused. Data URIs go through media_capture.split_inline, which bounds the length before it slices.

Notes for other lanes

  • Python auto (feat(python): one shared attachment manager in the auto module (traceai) #692): each package exports <name>_attachment_capabilities for _MEDIA_ADAPTERS; this now includes traceai_vllm and traceai_minimax. Leave out traceai_deepseek, traceai_cerebras and traceai_together: their descriptors have no operations.
  • Behaviour change, attachments on only: with attachments enabled for traceai_xai / traceai_minimax and traceai_openai enabled, a call has one span (the OpenAI instrumentor's) instead of two. With attachments off, both spans stay in either setup order (CI row test_binding_without_attachments_keeps_its_span); a canonical span diff against base is still owed.
  • feat(python): framework media capture, one owner per model call (LangChain, LlamaIndex, Agno, Pydantic AI) #691 (frameworks): it adds its own attach_bytes to media_capture. This PR keeps its bytes path inside provider_media, so the two do not conflict textually. The reconciliation should keep one.

Review fix pass (head 774675b)

Fixes from the independent review of 73727e7:

  1. LiteLLM PDF parts. LiteLLM claims the call, which used to drop file / input_file PDF parts the nested OpenAI instrumentor would otherwise capture. LiteLLM now captures them itself (compatible_messages(..., documents=True), LiteLLM only). Reviewer probe: base 1 upload, previous head 0, now 1 (completion, document).
  2. xAI + OpenAI on one manager. traceai_xai defers to an enabled traceai_openai, as Fireworks does. Reviewer probe in both setup orders: previous head 2 uploads, now 1 (ChatCompletion).
  3. M12 vLLM and MiniMax. Thin bindings, descriptors and fixture rows (see the matrix). MiniMax skips media while traceai_openai is enabled; a test proves one upload in both orders.
  4. Vertex descriptor. Advertises only the proven sync PredictionServiceClient paths; GenerativeModel / ImageGenerationModel are listed as unsupported until they have fixtures.
  5. sniff. Checks the ISO-BMFF major brand: HEIC/AVIF/QuickTime/M4A labelled video/mp4 is refused.
  6. Data URIs. Parsed with media_capture.split_inline, so the length is bounded before any slice.
  7. InvokeModel tool_result. Children carry the block index p, as Converse does.

Deferral gating and deny propagation (head d058417)

Binding decisions applied to traceai_xai and traceai_minimax:

  1. Deferral only with attachments on. The whole call goes to an enabled traceai_openai only while attachments are enabled for the binding (attachments= / attachment_manager= resolved to an active manager). With attachments off, the code path is base's: the binding span and the OpenAI span both stay. This restores base tracing for xAI (774675b had dropped the xAI span even with attachments off). MiniMax now defers the whole call instead of keeping its span and skipping media.
  2. Local deny wins downstream. A deferred call runs under provider_media.deferred(self._config) (new, see Privacy defaults). A binding with hide_input_images, an input hide flag or pii_redaction therefore uploads nothing, even when traceai_openai on the same manager uses a default TraceConfig. Reviewer probe local-deny-openai-default on 774675b: payloads=1.
  3. Context kept. A per-call policy and an outer owner mark that are already in the context still apply after the deferral (merged, never overwritten). If the helper cannot be loaded, the call runs under a deny-all policy.

New rows: test_binding_without_attachments_keeps_its_span (both orders, sync + async), test_deferred_call_keeps_binding_local_deny (no flag / hide_input_images / hide_inputs / pii_redaction, sync + async), test_deferred_call_keeps_context_policy_and_owner, plus core unit rows for deferral_policy / deferred. Job minimums were raised: core 54, xai 12, minimax 13.

CI on d058417: 48/48 checks pass. Passed rows: core 54, xai 12 (floor and latest), minimax 13 (floor and latest).

The private-VM qualification owed here was run on head 11e6b91 (next section).

Mixed setup fix and VM proof (head 11e6b91)

Review P2: _defers() only checked that traceai_openai was enabled. With the binding holding a manager and OpenAIInstrumentor enabled without one, the call was handed to an instrumentor that captures nothing: the binding span and its media were both lost (0 uploads).

  1. Accessor. OpenAIInstrumentor.captures_attachments() is true only while it is instrumented with an active shared manager (the same _attachment_manager resolution it already used). It imports nothing.
  2. Deferral needs both sides. traceai_xai / traceai_minimax defer only when they have a manager AND traceai_openai captures. Otherwise they keep the call, trace it and capture its media themselves. An older traceai_openai without the accessor counts as not capturing.
  3. Tests. test_binding_keeps_call_when_openai_has_no_manager (both orders, sync + async: one upload per call, owned by the binding span, both spans kept); test_binding_without_manager_keeps_its_span_next_to_capturing_openai (both orders: binding span kept, one upload owned by ChatCompletion); test_older_openai_without_accessor_never_takes_the_call. Job minimums: xai 17, minimax 18.

CI on 11e6b91: all checks pass; the new rows ran (xai 17 floor/latest, minimax 18 floor/latest, core 54).

Private VM on 11e6b91 (synthetic, loopback only, outbound network disabled):

  • All 33 CI matrix rows reproduced with the workflow's install and test mapping, each at or above its minimum.
  • Reviewer probe file, unmodified, 774675b vs this head (floor and latest):
    • local-deny-openai-default: payloads 1 -> 0 (xai and minimax);
    • openai-no-manager: payloads 0 -> 1, owned by the binding span, both spans kept (the P2 fix);
    • xai tracing-only: spans [ChatCompletion] -> [ChatCompletion, xAI Chat Completions] (base tracing restored).
  • Attachments-off canonical span dump (sync, async, stream; both setup orders) for xai and minimax, floor and latest: this head is IDENTICAL to base f5e77e0 in all 8 cells (base vs base re-run also identical). 774675b differed for xai (the dropped binding span).
  • Package suites, head vs base: xai 30/30 both; minimax 26/26 both; openai 1845 passed + 9 skipped (openai 3.49.0) and 1821 passed + 33 skipped (1.69.0), identical on both.
  • Mutations: 16/16 detected (deferral ungated, deferral removed, deferred context dropped, async path unwrapped, each for xai and minimax; context policy overwritten instead of merged, core and compat; hide_input_images not folded, core and compat; pii_redaction not folded; deferral back to "OpenAI enabled" for xai and minimax; accessor ignoring the manager).
  • ruff F/E9 on the changed files: only 3 unused imports in traceai_minimax/_wrappers.py, all present on base.

Testing

All runs are synthetic. The real installed SDKs talk to a loopback provider server and a TLS loopback F01 attachment server, with outbound network disabled (unshare --net). No real-provider call was made; real-provider evidence (Q04) is still owed.

Fix pass, private VM, every CI row reproduced with the workflow's exact install and test mapping (floor on Python 3.10, latest on 3.12). 31 rows ran on b68e065; the commits after it change tests only, and the 4 affected rows (bedrock and litellm, floor and latest) were re-run on 774675b. CI on 774675b is the full run of the final head. All 33 rows pass at or above their minimum passed counts: core 45, google-genai 9, vertexai 5, bedrock 12, litellm 9 floor / 10 latest, groq 2, together 3, fireworks 3, xai 5, vllm 4, minimax 6, deepseek 3, cerebras 3, mistralai 5, openrouter 3, portkey 3, ollama 4.

  • litellm 1.53.1 rejects chat file parts before sending, so the PDF row skips there (the floor minimum stays 9) and runs on 1.109.1.
  • Package suites, head vs 73727e7, identical: xai 30/30, vllm 117/117, minimax 26/26, litellm 201/201, vertexai 26 passed + 1 skipped, bedrock 41/41.
  • Mutations, 9/9 detected: brand check removed; data-URI slicing before the bound; part_index back to q; LiteLLM documents=True dropped; file parts not opt-in; xAI deferral removed; MiniMax OpenAI skip removed; vLLM capture call removed; MiniMax local TraceConfig ignored.

CI (earlier run of record: head 73727e7). All 44 checks pass, including the 29 new "Provider media" jobs and the #690 "Provider attachments" floor/latest jobs.

  • Every job must reach a minimum number of passed rows. The first run on 862548d was green but hollow for the compatible-chat jobs: a file plus a node ID in the same file collapsed to the node ID, so only the descriptor rows ran. That was fixed in af66415.
  • Passed rows per job: core 27, google-genai 9, vertexai 5, bedrock 12, litellm 9, groq 2, together 3, fireworks/xai/deepseek/cerebras 3 each, mistralai 5, openrouter 3, portkey 3, ollama 4.

Private VM, same suites at floor (Python 3.10) and latest (Python 3.12):

Package Floor Latest
google-genai 1.8.0 3.2.0
google-cloud-aiplatform 1.61.0 1.101.0
boto3 1.38.17 1.44.1
litellm 1.53.1 (earliest on PyPI ≥ the 1.43.0 floor) 1.109.1
groq 0.9.0 1.7.0
together 1.3.0 3.1.1
openai (fireworks/xai/deepseek) 1.69.0 3.49.0
cerebras-cloud-sdk 1.0.0 1.91.0
mistralai 1.1.0 (wrapper path) 3.7.0 (native-hook path)
openrouter 1.3.0 1.3.107
portkey-ai 1.9.0 2.3.4
ollama 0.3.0 0.6.3

Regression check: each touched package's own test suite, run on head and on base (f5e77e0):

  • Failure sets are identical for every package and tier.
  • Pre-existing failures, identical on base:
    • ollama 0.3.0: 32;
    • fireworks with openai 3.49.0: 5 collection errors;
    • google-genai 3.2.0: 1 collection error (httpx missing).
  • The VM caught a LiteLLM regression: its suite calls the wrappers with a stand-in self, which gave 45 failures. It was fixed in d0ea4e8 (now 201/201 on both tiers). CI has no LiteLLM package workflow.

Mutations: 19/19 detected. Each mutation was applied to the final tree and run against the latest venvs; the mistralai 1.x wrapper mutation used the 1.1.0 venv.

  • Helper:
    • magic-byte check removed;
    • local policy ignored;
    • data-URI gate made exact;
    • URL read before preflight;
    • untyped sources not limited to images;
    • per-observation part cap removed.
  • Google: generated output dropped; Imagen output dropped.
  • Bedrock:
    • Converse model allowlist opened;
    • computer tool ignored;
    • video allowed for any model;
    • Titan output dropped.
  • LiteLLM: owner claim removed (the nested OpenAI instrumentor then uploads a second copy); generated images dropped.
  • Groq: audio allowed in chat.
  • Mistral: native and 1.x capture dropped.
  • Ollama: generate images dropped.
  • OpenRouter: capture dropped.

Owed:

  • Q04 real-provider evidence for each advertised route.
  • Vertex: GenerativeModel / ImageGenerationModel fixtures (they go through the same gapic wrap, but the descriptor does not claim them until tested), the async client and streamed output.
  • vLLM and MiniMax real-server evidence (loopback fixtures only so far).
  • Gemini generate_images / generate_videos (not wrapped).
  • Together chat (pre-existing: no chat wrapper is installed).

🤖 Generated with Claude Code

nik13 and others added 11 commits November 11, 2026 19:04
Byte-identical to the media_capture.py change on feat/attachments-python-auto
(1b3072b + 0b2eb4b), so merging that branch later is a no-op here.
Provider adapters read CAPTURE_POLICY_KEY / OWNER_KEY through
local_capture_policy and Observation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…LM and compatible chat

New fi_instrumentation.attachments.provider_media (mirrors the TypeScript
provider-media helper): preflight before any read, exact-type re-admission,
magic-byte check after admission, deny-wins policy through media_capture,
and a per-package capability table that is the Python support matrix.

Bindings: google-genai (generate_content in/out, stream input), vertexai
(GenerateContent in/out, Imagen predictions), bedrock (Converse incl. tool
screenshots, InvokeModel Anthropic/Nova/Titan/Nova Canvas), litellm
(completion input, image_generation output, owner claim around the
downstream SDK call), groq/together/fireworks/xai/mistralai/openrouter/
portkey/ollama chat media. deepseek and cerebras export an empty
descriptor (no qualified media surface).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…a; CI workflow

Vertex Imagen predictions are a proto repeated field that does not slice,
so the bounded walk uses islice. traceai_together installs no chat wrapper
today, so its descriptor claims nothing. New workflow runs every provider
binding at its declared floor and latest SDK with outbound network off.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The LiteLLM package suite calls the wrapper methods with a stand-in self,
so the media helpers are module functions that read the manager with
getattr (no manager: unchanged behavior).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A file plus a node ID in the same file collapsed to the node ID, so the
compatible-chat jobs only ran the descriptor rows. Each job now also
requires its minimum number of passed rows, and the step runs with
pipefail.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ap is exact

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…wner, vLLM/MiniMax bindings, Vertex descriptor, MP4 brand, data-URI bound, InvokeModel part index)

- LiteLLM claims the call from a nested OpenAI instrumentor, so it now captures
  PDF file/input_file data URIs itself (documents=True, LiteLLM only): one upload.
- traceai_xai defers to an enabled traceai_openai like fireworks: one owner per call.
- Thin vLLM (OpenAI dialect) and MiniMax (image dialect) bindings with descriptors;
  MiniMax skips media while traceai_openai is enabled.
- The vertexai descriptor advertises the proven PredictionServiceClient paths only.
- sniff() accepts video/mp4 only for MP4 ISO-BMFF major brands (not HEIC/AVIF/QT/M4A).
- Data URIs go through media_capture.split_inline: bounded before slicing.
- Bedrock InvokeModel tool_result children carry the block index, as Converse does.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ip LiteLLM releases that reject file parts

The span ref carries message and part indexes only. litellm 1.53.x rejects chat
file parts before sending, so the floor job proves the PDF row on releases that
accept them (latest) and keeps its 9-row minimum.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…r message)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nik13 and others added 2 commits November 12, 2026 05:58
…local deny

traceai_xai and traceai_minimax now hand the whole chat call to an enabled
traceai_openai only while attachments are enabled for them. With attachments
off they trace exactly as before (binding span plus the OpenAI span).

A deferred call runs under provider_media.deferred(config): the binding's
local hide/PII flags are folded into a restriction-only CapturePolicy and
merged (deny wins) with the per-call policy under CAPTURE_POLICY_KEY, so the
OpenAI instrumentor cannot capture more than the binding allows, whatever its
own TraceConfig. hide_input_images denies every input part of the deferred
call (a CapturePolicy cannot single out images).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nik13 added a commit that referenced this pull request Nov 12, 2026
Cherry-picked from e158682 (feat/attachments-python-providers, #697). Every
non-docstring hunk of media_capture.py is byte-identical to that commit (and to
feat/attachments-python-auto), so merging those branches later is a no-op here.
The docstring keeps both bullets; the next commit rewrites the per-span one.

(cherry picked from commit e158682)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nik13 and others added 2 commits November 12, 2026 12:47
…ptures

traceai_xai and traceai_minimax handed the whole call to traceai_openai
whenever it was enabled. In a mixed setup (binding with an attachment
manager, OpenAIInstrumentor enabled without one) that dropped the binding
span and its media: nothing was uploaded.

OpenAIInstrumentor gains a small public accessor, captures_attachments(),
true only while it is instrumented with an active shared manager. The
bindings now defer only when both sides capture; otherwise they keep the
call, trace it and capture its media themselves. An older traceai_openai
without the accessor is treated as not capturing.

Tests: mixed setup in both orders (sync and async) uploads each image once,
owned by the binding span, with both spans kept; legacy-accessor stub never
takes the call. CI minimums: xai 15, minimax 16.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… has no manager)

With traceai_openai capturing and xAI/MiniMax instrumented without a
manager, the binding keeps its own span (tracing unchanged) and the image
is uploaded once, owned by the OpenAI span. CI minimums: xai 17, minimax 18.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant