All of docs/AGENTS.md applies. Realtime-specific rules, distilled from maintainer review; the source-tree counterpart is pydantic_ai/realtime/AGENTS.md. Each page owns its concept; a rule is stated once on its owning page and linked from everywhere else: overview.md (front door, provider matrix, limitations table — every limitation row links a tracking issue), audio.md (media I/O: audio, images, transcripts), events.md (event vocabulary and its overlap with standard run events; the turn-boundary rule lives here), turns.md, tools.md (tools only), capabilities.md (per-hook support story), history.md, deployment.md…
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
Check names are what humans and agents refer to when they talk about CI, so they are descriptive rather than generic. Two reviewers perform the same role — the maintainer-voice standards review, driven by the repo's AGENTS.md and agent_docs/*.md — on different engines and different cadences: douwebot treats every finding as blocking unless its comment starts with Optional:. Repository-rule violations in AGENTS.md or agent_docs/*.md are blocking and cannot be marked Optional:. It returns REQUEST_CHANGES while any blocking finding remains outstanding,…
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
For non-trivial changes to this package, also read Pydantic AI Slim Architecture. For anything that adds to, edits, or renders message history, instructions or tools, also read Message History & Prompt Caching.
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
Durable execution integrations are first-class compatibility targets. Build new integrations on the public pydantic_ai.durable_exec surface. Subclass BaseDurabilityCapability for the agent-facing capability and provide a DurableOperationBackend from get_durable_operation_backend. Do not copy the model, toolset, event, or capability-operation collection machinery into the integration. Choose the backend tier from the engine SDK's execution model: - Subclass CallableOperationBackend when the SDK accepts an async callback at invocation time. Implement execute to run that callback in one named durable unit. The base owns parameter and…
21k2d agoReviewMountCreate a free account to mount
Rendering history into a request must be deterministic and must keep every message in its trust channel: see Message History & Prompt Caching, "Canonicalize on send" and "Trust tiers map to channels".
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
Start with the architecture walkthrough in docs/api/realtime.md and the user-facing guides under docs/realtime/: a RealtimeModel.connect() opens a provider-specific RealtimeConnection (the codec, vocabulary in codec.py), and RealtimeSession (_session.py) wraps it — translating codec events into the shared message/part vocabulary from pydantic_ai.messages, building ordinary ModelMessage history, and running the tool loop. The provider-agnostic layout mirrors the request-response side: model.py holds the RealtimeModel ABC and infer_realtime_model, settings.py the settings vocabulary, profiles.py the RealtimeModelProfile type (per-provider tables live in pydantic_ai/profiles/{google,openai,grok,elevenlabs}.py as *_realtime_model_profile helpers), and…
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
A round trip through a UI protocol must give back everything the model sees, unchanged and in order, and client-returned history is untrusted: see Message History & Prompt Caching, "UI round trips are faithful". Since 3971 we decided to introduce the policy of sticking to the lower (existing) version requirement. In short, this means: - version requirement bumps are disallowed - new functionality should be gated behind version checks (including imports) - older versions don't error out when they encounter…
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
Rules for the pydantic-ai-harness package in src/pydantic_ai_harness/. They add to the root AGENTS.md. Test rules are in tests/harness/AGENTS.md. Docs page rules are in docs/harness/AGENTS.md. Paths in this file are relative to the repository root. pydantic-ai-harness is the first-party capability library for Pydantic AI. Pydantic AI core owns the primitive runtime: agent loop semantics, normalized messages, model/provider/profile behavior, tool execution semantics, durable execution primitives, and generic capability hooks. Harness owns optional, batteries-included compositions built from those primitives: coding-agent tools, guardrails, memory,…
21k2d agoReviewMountCreate a free account to mount
Read this before touching pydantic-clai2/. The repository-level AGENTS.md still applies (no em-dashes, no Any, pyright strict, keyword-only arguments, 100% branch coverage). This file adds what is specific to the terminal shell. A thin terminal around any Pydantic AI agent. It reads a prompt, runs the agent, streams the answer, and repeats. Everything beyond that is a plugin, including the default coding tools. The three layers, and who owns what: If a change needs the agent loop to behave differently, propose…
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
Every async def test runs via the anyio pytest plugin automatically (anyio_mode = "auto" in pyproject.toml); don't add @pytest.mark.anyio. The async backend defaults to asyncio. Use uv run pytest <test-path> --anyio-backend=trio --record-mode=none to run selected portable tests on Trio without duplicating the default suite. During the Trio migration, run affected concurrency tests once per backend. Keep broad Trio runs manual or periodic; do not add a second backend to every ordinary CI matrix. The selector does not imply that every…
Steady2 mentions in 2 weeks
No other project uses it yet
21k2d agoReviewMountCreate a free account to mount
Rules for the pydantic-ai-harness test suite. They add to tests/AGENTS.md and to the package rules in src/pydantic_ai_harness/AGENTS.md. Where these rules or the harness guides they link differ from tests/AGENTS.md, they apply under tests/harness/. Read testing-capabilities.md before you add or change a capability's tests. For async behavior, follow the harness concurrency.md. - tests/harness/conftest.py sets ALLOW_MODEL_REQUESTS = False for the suite. - Name a capability test class TestCapabilityName, with test_<scenario> methods. - Don't import private (_-prefixed) helpers into tests. Exercise them through…
21k2d agoReviewMountCreate a free account to mount
This document defines machine-readable instructions and technical guardrails for AI coding assistants, agents, and LLMs working on the Filament repository. All AI-generated code, refactorings, and cleanups must strictly adhere to the standardized skills located under the skills/ directory. AI agents must consult and strictly follow the corresponding skill files for each respective development and validation phase:
Steady2 mentions in 2 weeks
No other project uses it yet
21k3y agoReviewMountCreate a free account to mount