NeMo Relay 0.5.0

We are proud to release NeMo Relay 0.5.0!

Key Features

• 🔌 Dynamic plugin platform: plugin discovery, trust/attestation policies, lifecycle controls, and schema-aware plugins.toml configuration.
• 🧩 Plugin authoring SDKs: native Rust rust_dynamic plugin SDK plus grpc-v1 worker support with Rust and Python SDKs.
• 🛠️ CLI setup experience: verified installers, plugin configuration workflows, and expanded readiness diagnostics via nemo-relay doctor.
• 📊 Observability and runtime contracts: canonical LLM/tool intercept outcomes with ordered pending marks, plus stable normalized token and model-cost semantics.
• 🧬 Better LLM lineage: automatic Dynamo session and parent-session headers, with CLI turn-scoped trace correlation.
• 🌐 Binding and storage additions: Go PII-redaction helpers, ATIF storage helpers, model-pricing helpers, and broader adaptive-runtime binding parity.

Curated Breaking Changes

• Plugin configuration migration: plugins.toml is now the source of truth
• *Intercept contract change: LLM request and tool execution intercepts must return canonical outcome types, including pending marks.
• Removed surfaces and integrations: the experimental WebAssembly package, Cursor CLI support/entry points, and patch-based integrations are removed.
• Model pricing rename: pricing-related public surfaces are renamed to model pricing.

Non-Exhaustive Fixes

• Runtime and packaging reliability: enabled required Tokio/runtime features in distributed builds; generated references correctly for published packages; resolved the PyO3 advisory and remediated vulnerable dependencies.
• Observability and export delivery: hardened ATOF HTTP export, created missing JSONL output directories automatically, preserved late marks in parent traces, and flushes gateway events during shutdown.
• CLI diagnostics and host integration: improved Codex plugin discovery, cross-platform hook paths, guided setup, and status reporting for daemon binding, exporters, and remote ATIF storage.
• Safety and correctness: guarded FFI request-intercept output pointers and restricted unsafe BestEffortAnyCodec pickle decoding.

What's Changed

💥 Breaking Changes

• fix!: rename pricing surfaces to model pricing by @willkill07 in #335
• fix!: drop Cursor support by @willkill07 in #333
• chore!: remove patch-based integrations by @willkill07 in #332
• refactor!: remove WebAssembly binding by @willkill07 in #339
• feat(plugin)!: support pending marks from LLM intercepts by @bbednarski9 in #327
• feat(plugin)!: support pending marks from tool execution intercepts by @bbednarski9 in #350
• fix!: streamline plugin configuration by @willkill07 in #346

✨ New Features

• feat: add Go PII redaction helpers by @willkill07 in #275
• feat: add ATIF storage binding helpers by @willkill07 in #274
• feat: add pricing plugin helper bindings by @willkill07 in #273
• feat: add adaptive runtime binding parity by @willkill07 in #272
• feat: add dynamic plugin control-plane foundation by @afourniernv in #279
• feat: support dynamic plugin discovery via plugins.toml by @afourniernv in #290
• feat: add dynamic plugin lifecycle CLI commands by @afourniernv in #292
• feat: implement dynamic plugin host policy and attestation gates by @afourniernv in #302
• feat(types): add shared relay DTO crate by @willkill07 in #305
• feat(plugin): add native Rust plugin SDK by @willkill07 in #306
• feat(plugin): load rust_dynamic plugins in relay by @willkill07 in #307
• feat(worker): add grpc-v1 protocol and Rust worker SDK by @willkill07 in #308
• feat(worker): activate grpc-v1 worker plugins in relay by @willkill07 in #309
• feat(python): add nemo-relay-plugin worker SDK package by @willkill07 in #310
• feat: configure dynamic plugins through CLI by @willkill07 in #325
• feat: add built-in Dynamo session lineage by @willkill07 in #348
• feat: add verified CLI installers by @bbednarski9 in #354
• feat: add setup readiness diagnostics by @willkill07 in #357

⚡ Improvements

• perf: snapshot middleware callbacks before execution by @fallintoplace in #283
• refactor: hoist provider schema detection and normalization by @zhongxuanwang-nv in #291
• refactor: align exporter cost fallback with normalized responses by @mnajafian-nv in #300
• refactor: codec-owned provider-surface detection via a built-in registry by @zhongxuanwang-nv in #301
• refactor: prefer normalized exporter fallbacks by @mnajafian-nv in #304
• refactor: share agent and provider extraction strategies by @yczhang-nv in #318
• refactor: use reqwest rustls provider defaults by @willkill07 in #352

🐛 Bug Fixes

• fix: guard FFI request intercept output pointers by @fallintoplace in #284
• fix: enable tokio time for core package builds by @willkill07 in #285
• fix: harden ATOF HTTP endpoint export by @bbednarski9 in #295
• fix: keep late mark spans in parent traces by @bbednarski9 in #296
• fix(atof): create ATOF output directory before opening JSONL output file by @willkill07 in #297
• fix(cli): parse codex plugin list text output instead of --json by @zhongxuanwang-nv in #314
• fix(cli): use forward slashes in transparent hook executable path by @nanzhijin in #320
• fix: make native dynamic plugin example satisfy trust policy by @bbednarski9 in #326
• fix: enable runtime features in distributed packages by @willkill07 in #331
• fix: implement worker invocation cancellation by @willkill07 in #336
• fix: manage Python dynamic plugin environments by @willkill07 in #337
• fix: flush gateway events on shutdown by @willkill07 in #347
• fix: restrict BestEffortAnyCodec pickle decoding by @willkill07 in #344
• fix: report daemon bind and exporter status by @willkill07 in #345
• fix: report ATIF remote storage destinations in CLI status banner by @zhongxuanwang-nv in #349
• fix(security): resolve PyO3 advisory; temporarily ignore object_store advisory by @willkill07 in #351
• fix(cli): guide setup into plugin configuration by @ericevans-nv in #358
• fix: generate references for published packages by @willkill07 in #378

✅ Tests

• test: improve Go and FFI coverage by @willkill07 in #298
• test: fix ResolvedConfig initializer in launcher startup test by @zhongxuanwang-nv in #299
• test: stabilize dynamic worker cleanup lock test by @willkill07 in #353
• test: backfill extraction and telemetry parity coverage by @yczhang-nv in #355

🧰 Maintenance

• chore: prepare 0.4 code freeze by @willkill07 in #250
• ci: make release publishing resumable by @willkill07 in #269
• ci: stop release 0.4 nightly tags by @willkill07 in #271
• chore: address current Sonar findings by @willkill07 in #276
• ci: publish pii redaction crate from GitLab by @willkill07 in #278
• ci: comment license diffs on PRs by @willkill07 in #286
• chore: bump node tar and python protobuf by @willkill07 in #316
• chore: update Fern CLI to 5.57.0 by @willkill07 in #328
• ci: use sticky comments for license and docs updates by @willkill07 in #329
• chore(skills): add dynamic plugin maintainer guidance by @willkill07 in #312
• chore: remediate vulnerable dependencies by @willkill07 in #342
• ci: publish Python plugin wheel by @willkill07 in #356
• chore: add NAT reviewers for docs by @willkill07 in #384

📝 Documentation Updates

• docs: include PII redaction Rust API reference by @willkill07 in #280
• docs: clarify README quick start paths by @exactlyallan in #289
• docs: clarify runtime contracts and integration boundaries by @mnajafian-nv in #294
• docs: strengthen example coverage across runtime and integrations by @mnajafian-nv in #303
• docs: document LLM request and tool execution outcomes by @bbednarski9 in #341
• docs: require issues before contribution PRs by @willkill07 in #343
• docs: clarify Codex --config flag placement by @dagardner-nv in #338
• docs: document and stabilize token/cost field semantics by @zhongxuanwang-nv in #330
• docs: align package READMEs by @willkill07 in #377
• docs: reorganize plugin documentation by @willkill07 in #361
• docs: update 0.5 release notes by @willkill07 in #362
• docs: update stale plugin links by @willkill07 in #387

New Contributors

• @exactlyallan made their first contribution in #289
• @nanzhijin made their first contribution in #320
• @ericevans-nv made their first contribution in #358

Full Changelog: 0.4.0...0.5.0