v0.7.0

What's Changed

New Features

• feat: implement release-branch workflow by @ajbozarth in #1076
• feat: add health check endpoint to CLI serve by @markstur in #1100
• feat: add stopping sequence to model options by @psschwei in #1112
• fix: change hallucination detection tests to accept differences in explanation by @jakelorocco in #1059
• refactor: use plural logs to match otel by @AngeloDanducci in #1160
• feat: consolidate to llguidance from xgrammar by @jakelorocco in #1077
• feat: sandbox runtime and capability policy by @AngeloDanducci in #1171
• feat: groundedness requirement by @akihikokuroda in #773
• feat: python requirements by @akihikokuroda in #1128
• feat(intrinsics): introduce Adapter/Identity/IOContract/WeightsBinding scaffolding (Epic #929 Phase 0) by @planetf1 in #1158
• refactor: set do_sample=True if a seed is set on HF backends by @psschwei in #1149
• feat: support concurrency in base sampling strategy by @jakelorocco in #1175
• feat: python code interpreter improvements by @AngeloDanducci in #1190
• feat(cli): Multimodal serve images by @markstur in #1184
• feat: add matplotlib requirements by @akihikokuroda in #1208
• refactor(telemetry)!: move backend tracing onto plugin/hook pattern by @ajbozarth in #1181
• feat: dispatch release event to mellea-contribs by @avinash2692 in #1179
• feat(telemetry): wire metrics plugins to generation_batch_* hooks by @ajbozarth in #1254
• refactor(telemetry): align metrics/pricing init with tracing pattern by @ajbozarth in #1252
• feat: re-export serve types from public mellea.* namespace by @markstur in #1243
• feat: add matplotlib requirement factory by @akihikokuroda in #1227
• refactor(adapters): decouple KNOWN_CAPABILITIES from catalog entry names (#1186) by @planetf1 in #1266
• refactor(backends): move generate_from_raw hook firing into Backend base class by @ajbozarth in #1264
• feat: create sampling preset for python code generation by @akihikokuroda in #1265
• feat(sampling): re-export hidden strategies and fix structured-output docs by @planetf1 in #1281
• feat: add debug plugin collections by @akihikokuroda in #1251
• feat: add shell tool by @akihikokuroda in #1107
• feat: allow images as urls for supported backends by @AngeloDanducci in #1260
• feat(stdlib): public error and generate_log surface on ModelOutputThunk by @ajbozarth in #1307
• refactor(telemetry): migrate application spans to plugin hooks by @ajbozarth in #1289
• feat(intrinsics): shim old adapters onto Adapter dataclass; call_intrinsic rewritten (Epic #929 Phase 1) by @planetf1 in #1269
• feat: model aware sliding window context by @AngeloDanducci in #1270
• refactor(formatters): remove unused Granite32/33 processor classes by @planetf1 in #1337
• refactor(core): consolidate backend raw responses onto mot.raw namespace by @ajbozarth in #1329
• feat: standardize logits in mot by @AngeloDanducci in #1261
• refactor(intrinsics): migrate rag.py to new Adapter types (Epic #929 Phase 1 Wave 3) by @planetf1 in #1321
• refactor(core): promote ModelOutputThunk.thinking and deprecate _thin… by @exequeryphil in #1315
• refactor(core): split MOT internal state into _call / _gen sub-objects by @ajbozarth in #1350
• refactor(intrinsics): guardian.py whole-file migration + documents= kw-only discovery (#1139) by @planetf1 in #1323
• feat: first pass at mot as a separate class by @AngeloDanducci in #1348
• refactor(intrinsics): requirement_check_to_bool raises on schema mismatch (Epic #929 Phase 1) by @planetf1 in #1320
• feat: first pass at guardian io contract subclasses by @AngeloDanducci in #1357
• feat: context compaction strategies for the react loop by @yelkurdi in #996
• feat: add safe repr implementations for watsonx and litellm backends by @cptnm3 in #1359
• refactor(telemetry): migrate stream_with_chunking telemetry to plugin hooks by @ajbozarth in #1361

Bug Fixes

• fix: guard hf dep for m alora command by @psschwei in #1102
• fix: prevent double-wrapping of image data URIs in OpenAI message conversion by @markstur in #1091
• fix: add ollama async timeout by @psschwei in #1101
• fix: preserve discriminated-union schema in tool conversion by @planetf1 in #1104
• fix(streaming): replace sleep gate with _started event for cancel tests by @planetf1 in #1146
• fix: propagate generate_from_raw exceptions in OllamaModelBackend by @planetf1 in #1163
• fix: intrinsic tests and add some safeguards for future adapters changes by @jakelorocco in #1078
• fix(backends): populate mot._thinking from vLLM 'reasoning' wire key in LiteLLMBackend by @planetf1 in #1169
• fix: query-clarification-positive test_rag by @jakelorocco in #1178
• fix(ollama): soft-fail on empty response in generate_from_raw by @planetf1 in #1161
• fix(intrinsics): pin catalogue entries to HF revision SHAs + deduplicate requirement_check (#1135) by @planetf1 in #1157
• fix(hf): filter model_options for apply_chat_template via Jinja AST allowlist by @planetf1 in #1168
• fix(telemetry): break helpers->core import cycle so metrics plugins register by @ajbozarth in #1180
• fix: skip DCO check during git merge by @ajbozarth in #1189
• fix(retrievers): propagate non-404 HTTP errors from embedding download loop by @planetf1 in #1200
• fix: remove drift detection from intrinsic tests and pin tests to SHAs by @jakelorocco in #1196
• fix(litellm): pass api_base and set enable_thinking for bool THINKING by @planetf1 in #1207
• fix(streaming): use public cancellation property by @wondr-wclabs in #1220
• fix: Bumps to transformers==5.0.0 by @nrfulton in #418
• fix(backends): route ModelOption.THINKING=bool to chat_template_kwargs for vLLM by @planetf1 in #1204
• fix(hf): pass return_dict=True to apply_chat_template by @ajbozarth in #1253
• fix: demote format= warning to INFO and fix grammar by @RahulJoshi-rx in #1278
• fix: address default execution tier security posture by @AngeloDanducci in #1271
• fix(sampling): migrate budget forcing to mot.generation.usage; drop _meta["usage"] writes by @ajbozarth in #1288
• fix: update tool examples to use melleatool wrapper by @AngeloDanducci in #1297
• fix: mark xfail test_groundedness_e2e_string_documents by @akihikokuroda in #1205
• fix: update pillow version to prevent deprecation warning by @AngeloDanducci in #1287
• fix(tracing+ollama): split streaming span test into fast mock + slow e2e; add 300s default timeout by @planetf1 in #1302
• fix: test model_ids catalog and improve missing-model errors by @AngeloDanducci in #1299
• fix: pick best device (CUDA → MPS → CPU) for LoRA model loading by @planetf1 in #1303
• fix: improve repair instruction by @akihikokuroda in #1309
• fix: type assertion and richdocument docstrings by @AngeloDanducci in #1311
• fix: convert remaining rst docstrings by @AngeloDanducci in #1244
• fix: add request timeouts and mask API key in repr by @planetf1 in #1262
• fix: inter-chunk stream timeout and download part guard (RFC) by @planetf1 in #1236
• fix(intrinsics): restore aLoRA offsets and fix groundedness parser by @planetf1 in #1292
• fix(intrinsics): deprecate check_context_relevance — Granite 4.0 only adapter by @planetf1 in #1339
• fix: contribution guide for backticks in docstring by @jakelorocco in #1326
• fix: use IAM roles for bedrock access by @nrfulton in #578
• fix: generate_only kwargs for huggingface by @jakelorocco in #1325
• fix: add Python code extraction from tool_calls in PythonCodeExtraction validator by @akihikokuroda in #1353
• fix: plugin scope to handle list parameters by @jakelorocco in #1355
• fix: improve repair feedback for LLM-as-a-Judge validations by @akihikokuroda in #1248
• fix: get tool_call info from json schema by @akihikokuroda in #1360
• fix: fix jinja2 bare environment by @AngeloDanducci in #1363
• fix(test): enable openai intrinsic e2e tests in nightlies by @avinash2692 in #1341
• fix: canonicalize adapter io config path by @AngeloDanducci in #1366
• fix(adapters): correct typo from IntriniscsCatalogEntry to IntrinsicsCatalogEntry by @cptnm3 in #1370
• fix: message role not validated at runtime by @AngeloDanducci in #1371
• fix: enforce https in webhook urls by @AngeloDanducci in #1362
• fix: add chmod and umask guard to interpreter by @AngeloDanducci in #1372
• fix: host validation for vllm server-detection request by @AngeloDanducci in #1375
• fix(ci): mint release-workflow writes with the mellea-auto-release App token by @ajbozarth in #1376
• fix: repair release/v0.7 full-suite test failures (telemetry plugin registration + broken aLora example) by @ajbozarth in #1378

Documentation

• docs: fixing docs contributor guide by @psschwei in #1103
• docs: add redirect callout from MCP server page to MCP tools by @ajbozarth in #1110
• docs(api): normalize source docstring backticks by @planetf1 in #1155
• docs: migrate Guardian documentation from deprecated GuardianCheck to Intrinsics API by @planetf1 in #935
• docs: consolidate test guidance into test/README.md by @planetf1 in #1167
• docs: clarify uv add requires an existing uv project by @SAY-5 in #1152
• docs: migrate from Mintlify to Docusaurus 3 by @planetf1 in #1174
• docs(docusaurus): suppress unreleased banner until first snapshot exists by @planetf1 in #1237
• docs(docusaurus): remove author from Last Updated footer by @Ayman-ing in #1257
• docs: update alora examples readme by @AngeloDanducci in #1259
• docs: update readme files in docs/examples by @akihikokuroda in #1228
• docs: clarify sampling raises by @skyswordw in #1214
• docs(docusaurus): move version dropdown to left navbar; shorten label to 'main' by @planetf1 in #1238
• docs: clarify MelleaSession vs functional.act() layering; add start_session() pin example by @planetf1 in #1283
• docs: migrate examples off deprecated interpreters to python tool by @AngeloDanducci in #1324
• docs: clarify format= value contract and model_validate_json pattern by @planetf1 in #1333
• docs: update docs/examples/tools/README.md by @akihikokuroda in #1328
• docs: fix stale 'main (unreleased)' reference in CONTRIBUTING_DOCS.md by @planetf1 in #1335
• docs: document PR review states by @ajbozarth in #1306
• docs: add python / matplotlib requirement example by @akihikokuroda in #1293
• docs(core): document _generate_from_context uncomputed-thunk contract by @ajbozarth in #1354
• docs: update contributing and agent guidelines by @AngeloDanducci in #1364

Other Changes

• chore: add dco to contributing by @psschwei in #1129
• ci: fix blocked dco check in merge queue by @psschwei in #1133
• chore: add local DCO check via pre-commit commit-msg hook by @ajbozarth in #1131
• test: add hook ordering regression coverage by @tatakaisun in #1113
• chore: disable intrinsic io.yaml overrides by @psschwei in #1156
• test(backends): fix vision Ollama tests failing in CI with 400 model does not support vision (#1185) by @planetf1 in #1188
• test(formatters): skip combo fixtures gracefully when Hub returns 429 by @planetf1 in #1195
• test: centralise HF Hub skip guard to boost CI resiliency against 429 errors by @planetf1 in #1199
• chore: add HF_TOKEN for authentication by @jakelorocco in #1202
• chore(backends): drop deprecated torch_dtype argument in LocalHFBackend by @planetf1 in #1231
• chore: add PEP 484 return-type annotations to mellea.stdlib public me… by @yehorcallmedai-maker in #1263
• ci(quality): retry ollama model pull on transient network failure by @planetf1 in #1234
• chore(terminology): adopt agreed Granite Switch / Mellea glossary (#1192) by @planetf1 in #1256
• chore(terminology): sweep docs/examples and Python docstrings (#1192 follow-up) by @planetf1 in #1322
• test: auto-retry ollama tests on transient ReadTimeout by @ajbozarth in #1369

New Contributors

• @tatakaisun made their first contribution in #1113
• @wondr-wclabs made their first contribution in #1220
• @Ayman-ing made their first contribution in #1257
• @skyswordw made their first contribution in #1214
• @yehorcallmedai-maker made their first contribution in #1263
• @RahulJoshi-rx made their first contribution in #1278
• @exequeryphil made their first contribution in #1315
• @cptnm3 made their first contribution in #1359

Full Changelog: v0.6.0...v0.7.0