v0.8.0

What's Changed

New Features

• feat: surface container artifacts via executionresult artifacts by @AngeloDanducci in #1384
• refactor: make json_util a private module by @AngeloDanducci in #1399
• feat: audio text to text support by @markstur in #1396
• feat: carry reasoning across turns with per-backend replay policy by @AngeloDanducci in #1374
• feat: imageblock and imgurblock ergonomics by @AngeloDanducci in #1377
• feat: add license header remediation to pre commit by @AngeloDanducci in #1419
• refactor(intrinsics): promote _DictContract to shared adapters module; collapse guardian.py's factuality contracts by @planetf1 in #1428
• feat(telemetry): emit execute_tool spans for tool calls by @ajbozarth in #1430
• refactor: extract shared Ollama eviction helper from conftest files by @AngeloDanducci in #1434
• refactor: consolidate duplicate import_optional into optional.py by @AngeloDanducci in #1438
• refactor(backends)!: narrow AdapterMixin verbs, centralize option resolution, add AdapterFunctionMetricsPlugin skeleton by @planetf1 in #1422
• feat: add support for audio-text-to-text with m serve CLI and include examples by @markstur in #1443
• refactor(backends): unify mot.raw.response shape across streaming and non-streaming by @cptnm3 in #1449
• feat(telemetry): add sampling and validator tracing spans via hooks by @ajbozarth in #1488
• refactor: enforce act/aact without strategy by @AngeloDanducci in #1468
• refactor: introduce alias for component cblock modeloutputthunk union by @AngeloDanducci in #1458
• feat!: emit per-chunk streaming span events on backend spans by @ajbozarth in #1496
• feat: add mellea version dunder by @jakelorocco in #1524
• feat(cli): allow model passed by OpenAI client to be used in the served Mellea program by @markstur in #1512
• feat(backends): add default_extra_body to OpenAIBackend by @planetf1 in #1536
• feat: remove m async fix command (and other improvements) by @jakelorocco in #1537
• feat: add nemotron model ids by @jakelorocco in #1526
• feat: allow components to declare their chat role via format_for_llm by @AngeloDanducci in #1542
• feat: add zizmor audit job to CI quality workflow by @planetf1 in #1530
• feat(backends): LocalFileBinding implements verbs (PEFT/aLoRA path) + from_catalog() (Epic #929 Phase 2) by @planetf1 in #1454
• feat: make call_tools public by @akihikokuroda in #1544
• feat(telemetry): align spans and metrics with OTel GenAI semantic conventions by @ajbozarth in #1551
• refactor(streaming)!: replace stream_with_chunking with single-task stream() by @ajbozarth in #1543
• refactor: improve message serialization by @AngeloDanducci in #1564
• refactor(intrinsics): resolve the adapter's output contract from the adapter, not a parallel argument by @planetf1 in #1556
• feat(backends): EmbeddedBinding implements apply_activation (Granite Switch path); remove render_controls + set_request_adapter (Epic #929 Phase 2) by @planetf1 in #1559
• feat: add author-pluggable extension point for provider-specific message fields by @AngeloDanducci in #1586
• feat(backends): populate mot.raw.response on the HF generate_from_raw by @cptnm3 in #1518
• feat: add Granite 4.2 model defaults by @planetf1 in #1587
• feat(backends): support embedded adapters on LocalHFBackend by @planetf1 in #1593
• feat(streaming): add EventStreamer for iterating stream events by @ajbozarth in #1567
• feat(backends): accept composed Adapter directly (Epic #929, issue #1144, PR1) by @planetf1 in #1619
• feat: add per-requirement chunking for streaming validation by @ajbozarth in #1630
• feat: preserve Context subtype and enforce input==output context type by @AngeloDanducci in #1582
• feat(core): audio file/URL constructors and representation-aware capability guards by @jakelorocco in #1601
• feat(backends): support Ollama adapter functions by @planetf1 in #1634

Bug Fixes

• fix(ci): serialize ollama requests to prevent llama-server segfault cascade by @ajbozarth in #1381
• fix: port telemetry test-order fix + broken aLora example skip to main (#1378) by @ajbozarth in #1382
• fix(ci): generate API/CLI docs before snapshotting release version by @ajbozarth in #1386
• fix: preserve tool_calls in multi-turn history and skip malformed tool-arg JSON for OpenAI-compatible backends by @mvanhorn in #1373
• fix(ci): revert OLLAMA_NUM_PARALLEL serialization (#1381) by @ajbozarth in #1387
• fix: test kv flake with rephrase and chatcontext by @AngeloDanducci in #1403
• fix: refactor bash validation by @akihikokuroda in #1400
• fix: force cpu accelerator for docling on apple silicon mps by @AngeloDanducci in #1417
• fix: ollama releases prior to the pre release today ran into issues with amx runners by @AngeloDanducci in #1404
• fix: pypi build files by @jakelorocco in #1414
• fix: updated assert to fix flakeyness of sampling example by @rphadnis in #1420
• fix: bump pinned Ollama version to 0.32.2 (rc0 tag removed) by @planetf1 in #1425
• fix: raise ValueError for image/audio inputs in LocalHFBackend by @markstur in #1410
• fix: Context.last_turn by @akihikokuroda in #1423
• fix: add recursive nested-model $ref resolution in Pydantic tool parameter schemas by @akihikokuroda in #1416
• fix: allow act to operate on CBlock and ModelOutputThunk by @AngeloDanducci in #1429
• fix: bump langchain-core and litellm floors to clear cve by @AngeloDanducci in #1447
• fix(hf): stop generation workers on stream timeout by @dk3yyyy in #1469
• fix(backends): merge intrinsic extra_body once by @binggao1230 in #1300
• fix: port analytics review improvements from website PR by @ajbozarth in #1463
• fix: doc autogen test assertions by @AngeloDanducci in #1480
• fix: change tool_calls from dict to list by @akihikokuroda in #1435
• fix(core): replace bare assert with ValueError in ImageBlock/AudioBlock validation by @markstur in #1490
• fix: assign the result of Context.add in transform/atransform by @planetf1 in #1477
• fix: do not share session state across requests in m_serve examples by @markstur in #1481
• fix: converge litellm to a single version across python versions by @AngeloDanducci in #1494
• fix(stdlib): resolve postponed annotations in generative stub signature by @planetf1 in #1504
• fix: surface HF auth failures when loading switch adapters by @psschwei in #1501
• fix: feed tool results to open ai compat providers as role tool with id by @AngeloDanducci in #1505
• fix: raise clear error when intrinsic context forwards no history by @psschwei in #1525
• fix: resolve postponed annotations without failing on TYPE_CHECKING return types by @planetf1 in #1509
• fix(hf): set whitespace_flexible=True in grammar_from_json_schema to prevent silent array collapse by @planetf1 in #1513
• fix: harden computed model output thunks by @jakelorocco in #1519
• fix: log non-OpenAI format= assumption once at init (#1502) by @sofiamartincraft in #1514
• fix: mellea top level imports to be more ergonomic and lazily import some things to make startup time faster by @jakelorocco in #1538
• fix(tools): stop validate_tool_arguments padding tool-call args with null by @avinash2692 in #1573
• fix: rewrite broken aLora 101 example for the intrinsics API by @planetf1 in #1557
• fix(backends): release() deregisters via a new remove_adapter() inverse verb by @planetf1 in #1554
• fix: force safe click version via uv constraint by @AngeloDanducci in #1581
• fix(backends): reject OpenAI model option overrides by @planetf1 in #1578
• fix: preserve optional status and default values in tool serialization by @AngeloDanducci in #1570
• fix(backends): route intrinsic generation through adapter_scope, fix lock reentrancy by @planetf1 in #1555
• fix(adapter-functions): remove deprecated context relevance intrinsic by @planetf1 in #1579
• fix(adapters): use effective capabilities in shim identities by @planetf1 in #1580
• fix: surface validate parse errors as ValidationResult.error instead of propagating by @AngeloDanducci in #1591
• fix: resolve HF symlinks in from_hub via temp local_dir by @Disha714 in #1495
• fix: validate_tool_arguments does not enforce Literal/const constraints on tool fields by @JessicaHe22 in #1547
• fix(hf): handle ModelOption.THINKING chat templates by @planetf1 in #1588
• fix(backends): serialize resolve_adapter() registration under the activation lock by @planetf1 in #1607
• fix(backends): fire adapter_function_invocation_complete for embedded adapter calls by @planetf1 in #1609
• fix(tools): preserve MCP server-side tool name after rename by @ajbozarth in #1612
• fix(rag): stop mirroring output schema key in groundedness batch prompt by @planetf1 in #1618
• fix(backends): deep-merge extra_body.chat_template_kwargs in merge_model_options by @planetf1 in #1617
• fix(hf): forward string THINKING levels as reasoning_effort, not low_effort by @planetf1 in #1639
• fix(rag): normalise model-returned JSON types by @planetf1 in #1625
• fix(telemetry): measure streaming latency at provider receipt by @ajbozarth in #1631
• fix(hf): parse tags into mot.thinking on LocalHFBackend by @planetf1 in #1616
• fix: close backend clients and event-loop threads to stop resource leaks by @psschwei in #1648
• fix: validation by @jakelorocco in #1628
• fix(test): reliable nightly GPU runs — phased execution as default, script-owned timings by @planetf1 in #1643

Documentation

• docs: sync changelog for v0.7.0 by @mellea-auto-release[bot] in #1383
• docs: complete RST to Google-style docstring sweep by @mvanhorn in #1397
• docs: add bash_executor example by @akihikokuroda in #1391
• docs: document safer execution patterns by @akihikokuroda in #1393
• docs: note policy_guardrails now raises AdapterSchemaMismatchError by @planetf1 in #1427
• docs: organize the m_serve examples in subdirectories by @markstur in #1459
• docs: add telemetry implementation guidance to AGENTS.md by @ajbozarth in #1461
• docs: use markdown backticks in docstrings by @Jonesxq in #1472
• docs: fix example notebook on colab by @psschwei in #1489
• docs(telemetry): document the public-API consumption boundary by @ajbozarth in #1541
• docs: add guidance on when to use which validation approach by @AngeloDanducci in #1540
• docs: delete docs/dev/, migrate current content, close #1464's telemetry-doc gap by @planetf1 in #1483
• docs: expand LiteLLM backend documentation by @psschwei in #1548
• docs: document construction-time ModelOption.THINKING suppression by @AngeloDanducci in #1571
• docs: document PR template and checklist bot in AGENTS.md by @AngeloDanducci in #1585
• docs: reframe docstring guidance as public API reference, not prompts by @ajbozarth in #1606
• docs: document PR template system by @psschwei in #1641
• docs: use relative .md links for all cross-doc references by @psschwei in #1647
• docs: document ModelOption.THINKING behaviour for Granite by @planetf1 in #1645

Other Changes

• chore: remove review guard on mellea/core by @psschwei in #1379
• chore: remove review guard on intrinsics by @psschwei in #1380
• chore: add spdx headers to code files by @psschwei in #1401
• ci: add check for license/copyright headers on code files by @psschwei in #1402
• chore: drop superfluous dco merge queue check by @psschwei in #1478
• test: add mocked backend tracing coverage by @dk3yyyy in #1471
• test: honor capability overrides during example collection by @snowyukitty in #1451
• chore: don't use nltk v3.10.1 by @psschwei in #1520
• test(vision): activate Ollama vision e2e via Hugging Face GGUF pull by @planetf1 in #1527
• test: assert const to enum schema survival for single-value Literal by @AngeloDanducci in #1596
• test(backends): add real-model GPU e2e for intrinsic adapter_scope routing by @planetf1 in #1608
• test(telemetry): avoid requiring provider chunk intervals by @planetf1 in #1626
• test: add integration-level test coverage for ModelOption.THINKING on OllamaModelBackend by @planetf1 in #1613
• test: add job for testing notebooks by @psschwei in #1649

New Contributors

• @mellea-auto-release[bot] made their first contribution in #1383
• @mvanhorn made their first contribution in #1373
• @rphadnis made their first contribution in #1420
• @dk3yyyy made their first contribution in #1471
• @Jonesxq made their first contribution in #1472
• @snowyukitty made their first contribution in #1451
• @binggao1230 made their first contribution in #1300
• @sofiamartincraft made their first contribution in #1514
• @Disha714 made their first contribution in #1495
• @JessicaHe22 made their first contribution in #1547

Full Changelog: v0.7.0...v0.8.0