Issues

Agent workflow provider failure: CI Review

pydantic/pydantic-ai#11008 · opened

on Oct 22, 2026

Agent workflow task failure: CI Review

pydantic/pydantic-ai#11005 · opened

on Oct 22, 2026

Agent workflow task failure: CI Review

pydantic/pydantic-ai#11001 · opened

on Oct 22, 2026

Feature request: Support offline evaluation of precomputed outputs without executing a task

pydantic/pydantic-ai#10997 · opened

on Oct 22, 2026

1

`thinking` is silently ignored for OpenAI GPT-5.6 and GPT-6 on Bedrock Converse

pydantic/pydantic-ai#10992 · opened

on Oct 22, 2026

1

`ctx.emit` raises in realtime sessions, so `StepPersistence` makes a session close raise

pydantic/pydantic-ai#10988 · opened

on Oct 22, 2026

1

AzureProvider sends $defs/$ref tool schemas to Grok, which rejects them

pydantic/pydantic-ai#10987 · opened

on Oct 22, 2026

1

Reduce cognitive complexity in provider adapters and gate new growth

pydantic/pydantic-ai#10983 · opened

on Oct 22, 2026

Realtime docs: a batch of wrong or over-general claims across the provider and concept pages

pydantic/pydantic-ai#10982 · opened

on Oct 22, 2026

1

Anthropic: unanswered `tool_search` call is dropped on the direct API, so the model repeats the search

pydantic/pydantic-ai#10966 · opened

on Oct 22, 2026

1

Support deterministic client lifecycle management in ImageGenerator

pydantic/pydantic-ai#10964 · opened

on Oct 22, 2026

1

Demoted tool_search replay rebuilds the regex pattern from an unescaped BM25 query

pydantic/pydantic-ai#10954 · opened

on Oct 21, 2026

1

`Agent.from_spec` ignores an explicit `output_type=str` when the spec has an `output_schema`

pydantic/pydantic-ai#10952 · opened

on Oct 21, 2026

1

Send message history to the OpenAI Decisions API as native input items once OpenAI confirms support

pydantic/pydantic-ai#10950 · opened

on Oct 21, 2026

Let a capability's tools emit capability events under `PrefixTools` and other renaming wrappers

pydantic/pydantic-ai#10927 · opened

on Oct 21, 2026

1

Approved tool runs again if the process dies after it executed but before the history was saved

pydantic/pydantic-ai#10926 · opened

on Oct 21, 2026

2

Each agent run stays in memory until a full garbage collection (two reference cycles)

pydantic/pydantic-ai#10924 · opened

on Oct 21, 2026

1

AnthropicModel raises `TypeError: Cannot instantiate typing.Union` with anthropic 2.0.0 (`BetaMessageParam` is now a union)

pydantic/pydantic-ai#10923 · opened

on Oct 21, 2026

1

Add a Boat.dev sandbox capability

pydantic/pydantic-ai#10922 · opened

on Oct 21, 2026

4

Gemini Live: a tool call without an id, lost across a resumed reconnect, ends the event stream with `ValueError`

pydantic/pydantic-ai#10912 · opened

on Oct 21, 2026

1

GoogleModel replays thought signatures across Vertex AI and Gemini API (both named 'google') → 400 Invalid thought signature

pydantic/pydantic-ai#10905 · opened

on Oct 21, 2026

3

HuggingFaceModel drops the `reasoning` field returned by the Inference API (non-streaming and streaming)

pydantic/pydantic-ai#10899 · opened

on Oct 21, 2026

1

`GoogleModel`: `IMAGE_RECITATION`, `IMAGE_OTHER` and `TOO_MANY_TOOL_CALLS` finish reasons are not mapped

pydantic/pydantic-ai#10898 · opened

on Oct 21, 2026

1

Docs: one-paragraph entry for AI Stupid Level (OpenAI-compatible router)?

pydantic/pydantic-ai#10895 · opened

on Oct 21, 2026

2

veltraaa

pydantic/pydantic-ai#10891 · opened

on Oct 21, 2026

1

Navigation Menu

Repository navigation

Issues

Agent workflow provider failure: CI Review

Agent workflow task failure: CI Review

Agent workflow task failure: CI Review

Feature request: Support offline evaluation of precomputed outputs without executing a task

`thinking` is silently ignored for OpenAI GPT-5.6 and GPT-6 on Bedrock Converse

`ctx.emit` raises in realtime sessions, so `StepPersistence` makes a session close raise

AzureProvider sends $defs/$ref tool schemas to Grok, which rejects them

Reduce cognitive complexity in provider adapters and gate new growth

Realtime docs: a batch of wrong or over-general claims across the provider and concept pages

Anthropic: unanswered `tool_search` call is dropped on the direct API, so the model repeats the search

Support deterministic client lifecycle management in ImageGenerator

Demoted tool_search replay rebuilds the regex pattern from an unescaped BM25 query

`Agent.from_spec` ignores an explicit `output_type=str` when the spec has an `output_schema`

Send message history to the OpenAI Decisions API as native input items once OpenAI confirms support

Let a capability's tools emit capability events under `PrefixTools` and other renaming wrappers

Approved tool runs again if the process dies after it executed but before the history was saved

Each agent run stays in memory until a full garbage collection (two reference cycles)

AnthropicModel raises `TypeError: Cannot instantiate typing.Union` with anthropic 2.0.0 (`BetaMessageParam` is now a union)

Add a Boat.dev sandbox capability

Gemini Live: a tool call without an id, lost across a resumed reconnect, ends the event stream with `ValueError`

GoogleModel replays thought signatures across Vertex AI and Gemini API (both named 'google') → 400 Invalid thought signature

HuggingFaceModel drops the `reasoning` field returned by the Inference API (non-streaming and streaming)

`GoogleModel`: `IMAGE_RECITATION`, `IMAGE_OTHER` and `TOO_MANY_TOOL_CALLS` finish reasons are not mapped

Docs: one-paragraph entry for AI Stupid Level (OpenAI-compatible router)?

veltraaa