v0.0.55 2025-04-09

What's Changed

• Allow empty user_prompt in run_stream by @lfloeer in #1421
• fix: infer azure provider and environment for key by @Wh1isper in #1422
• Fix bug with JSON schema generation when using InstrumentedModel by @dmontagu in #1424
• Add PydanticAI User-Agent header by @Kludex in #1425

New Contributors

• @lfloeer made their first contribution in #1421

Full Changelog: v0.0.54...v0.0.55

v0.0.54 2025-04-09

What's Changed

• evals: error when opentelemetry-sdk is not installed by @Kludex in #1403
• bedrock: allow empty system prompt by @Kludex in #1405
• fix import for the bedrock model by @yilololo in #1410
• Yield initial node during Graph (and therefore Agent) iteration by @dmontagu in #1412
• Make user_prompt optional by @Kludex in #1406
• Add stop_sequences to ModelSettings by @Kludex in #1419

New Contributors

• @yilololo made their first contribution in #1410

Full Changelog: v0.0.53...v0.0.54

v0.0.53 2025-04-07

What's Changed

• Improve some error messages by @dmontagu in #1366
• Improve serialization of LLMJudge and custom evaluators by @dmontagu in #1367
• Use handle_stream.stream_output() in the CLI by @Kludex in #1376
• Feature/add openai strict mode by @mscherrmann in #1304

New Contributors

• @mscherrmann made their first contribution in #1304

Full Changelog: v0.0.52...v0.0.53

v0.0.52 2025-04-03

What's Changed

• Add dependencies to evals by @Kludex in #1358

Full Changelog: v0.0.51...v0.0.52

v0.0.51 2025-04-03

What's Changed

• Match OpenAI models in strictness by @Kludex in #1356
• Switch mcp-run-python server to use deno by @samuelcolvin in #1340

Full Changelog: v0.0.50...v0.0.51

v0.0.50 2025-04-03

What's Changed

• Fix some issues with non-serializable inputs in evals by @dmontagu in #1333
• Add expected output to logfire attributes for the evals task span by @dmontagu in #1334
• Drop exclusiveMaximum/exclusiveMinimum from Gemini by @Kludex in #1341
• Drop JSON schema url from schema on Gemini by @Kludex in #1342
• Add py.typed marker to pydantic_evals package by @t94j0 in #1349

New Contributors

• @mike-luabase made their first contribution in #1331
• @t94j0 made their first contribution in #1349

Full Changelog: v0.0.49...v0.0.50

v0.0.49 2025-04-01

What's Changed

• Add Gemini 2.5 pro and improve CLI by @samuelcolvin in #1317
• Add OpenAI built-in tools by @Kludex in #1327
• Add generate_summary and truncation to OpenAIResponsesModelSettings by @Kludex in #1328
• Move imports to try-except block by @Kludex in #1329
• Don't download content on ImageUrl and DocumentUrl for Anthropic Models by @jnishiyama in #1318

New Contributors

• @jnishiyama made their first contribution in #1318

Full Changelog: v0.0.48...v0.0.49

v0.0.48 2025-03-31

What's Changed

• Add OpenAI Responses API by @Kludex in #1256
• Add dynamic version on evals extras by @Kludex in #1315

Full Changelog: v0.0.47...v0.0.48

v0.0.47 2025-03-31

What's Changed

• Ensure model settings' prefixes are enforced by @Kludex in #1257
• Bump logfire by @Kludex in #1262
• improvements while playing with the CLI by @samuelcolvin in #1263
• add span around tool calls by @samuelcolvin in #1158
• Add pydantic-evals package by @dmontagu in #935
• remove "preparing model request params" span by @samuelcolvin in #1281
• Clean up SpanQuery and related APIs by @dmontagu in #1287
• Fix the default types of generic params to EvaluatorContext by @dmontagu in #1290
• More otel cleanup for evals by @dmontagu in #1291
• Clean up public evals API by @dmontagu in #1298
• [Bugfix] Unknown tool called by agent will cause response 400 by @BeautyyuYanli in #1255
• feat: Add read timeout and connect timeout for bedrock provider by @Wh1isper in #1259
• allow str as model, fix Groq models by @samuelcolvin in #1306
• CLI fixes and improvements by @dmontagu in #1299
• Allow use of PYTHONOPTIMIZE=1 by @Kludex in #1307
• disable cohere with emscripten by @samuelcolvin in #1296

New Contributors

• @yaswhar made their first contribution in #1224
• @sarunas-llm made their first contribution in #1280
• @BeautyyuYanli made their first contribution in #1255

Full Changelog: v0.0.46...v0.0.47

v0.0.46

What's Changed

• Use different HTTP clients based on providers by @Kludex in #1242
• Add headers, timeout, and sse_read_timeout to MCPServerHTTP by @JohnUiterwyk in #1218
• Revert "Use get_running_loop instead of get_event_loop" by @Kludex in #1252

New Contributors

• @JohnUiterwyk made their first contribution in #1218

Full Changelog: v0.0.45...v0.0.46