Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
39 changes: 39 additions & 0 deletions release/notes/v0.4.0.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,39 @@
Lightspeed v0.4.0 adds scoped API keys and Platform roles, company SSO, a first-class runtime CLI, universe model defaults, standalone transcription, and more reliable long-running sessions. It also improves workspace file handling and adds manual context compaction to the web UI.

**Breaking changes and upgrade requirements**

- This is a coordinated breaking release. Runtime migrations have been consolidated and edited since v0.3.0, including replacement of migration 10. Although the required runtime schema revision remains 10, the migration names and checksums differ. The v0.4.0 migrator rejects an existing v0.3.0 ledger; there is no supported automatic in-place upgrade or conversion of the former identity/access model. Do not rewrite ledger rows to bypass validation.
- For retained installations, quiesce ingress, scheduled work, and workers; preserve a coordinated recovery set of PostgreSQL, Temporal, object storage, configuration, keys, and machine state, together with the v0.3.0 artifacts. Validate a conversion against isolated copies before using retained state with v0.4.0. Old session and workflow histories are not guaranteed compatible. Fresh installations use `lightspeed-runtime migrate` before starting workers. Disposable development installations can use `./dev.sh reset` and create fresh work; this is not a production migration procedure.
- Deploy matching runtime, Platform, connector, Configurator, CLI, and client artifacts. Platform now requires schema revision 4 and applies its own migrations at startup. Its migration coverage does not establish compatibility of an old runtime database. The API protocol remains `lightspeed.agent.api.v1` and the environment protocol remains 2; unchanged identifiers do not imply unchanged contracts or saved-state compatibility.
- The hosted executable is now `lightspeed-runtime`, replacing `lightspeed-server`; update service commands and archive references. The public TypeScript package is now `@lightspeed-ai/sdk`, replacing `@lightspeed-ai/agent-client`. Rust workspace packages `engine` and `temporal-server` are now `harness` and `temporal-runtime`. Source builds use Rust 1.99.0 and Node 24.21.0.
- Update authentication and API integrations for deployment/universe-scoped keys and method groups. Deployment operations move from `operator/*` to `deployment/*`; Platform owns memberships, roles, and access to unshared sessions. CLI resource commands and chat attachment flags have also changed; consult the current CLI reference when updating scripts.
- Remove `LIGHTSPEED_CHAT_PROVIDER` and `LIGHTSPEED_CHAT_MODEL`: either variable now prevents runtime startup. Configure each universe's **Agent runs** and **Speech-to-text** defaults separately. Explicit session and profile models take precedence; requests needing an unset default fail with `model_default_unset`. Provider credentials alone do not select a model.
- Audio must be transcribed before session admission. Run start, context append, and steering reject raw audio; use the standalone transcription API and submit prepared text.
- New sessions and configuration replacements default to automatic standalone compaction when the policy is omitted. Historical omitted policies retain disabled replay behavior; set Disabled explicitly when automatic compaction and context-limit recovery should stay off.
- Review installed environment daemons and use the matching release. Protocol-mismatch auto-upgrade will not install v0.4.0 solely because its product version is newer.

See [Upgrades and recovery](https://ls.bot/docs/deployment/upgrades-and-recovery/) for recovery planning and [the CLI guide](https://ls.bot/docs/using-lightspeed/cli/) for current commands.

**Access, identity, and administration**

- Scoped API keys support method-group restrictions, opaque actor attribution, revocation, and immediate secret rotation. Rotation preserves authority and invalidates the old secret without a grace period.
- Platform adds Viewer, Contributor, Operator, and Admin roles, private sessions and explicit sharing, member management, and feature controls. Configurator MCP exposes tools permitted by the caller's method groups.
- Company OIDC sign-in supports application/admin entitlements, bounded sessions, suspension, revocation, emergency local administrators, and an access audit trail. Universe memberships remain managed in Lightspeed. Existing local accounts are not linked to company accounts by email; designate and test emergency access before enabling SSO. SCIM and automatic directory deprovisioning are not included.

**CLI, models, and voice**

- Use saved authenticated connections, protected local credentials, universe selection by UUID or slug, and runtime administration without requiring Platform. Development startup authenticates by default and provides `lightspeed connect dev`.
- Chat gains recent-session resume/continue, runtime model discovery, compatible session-local model changes, restored transcript history, clearer tool activity, and optional usage/timing statistics. Provider identity and API kind remain fixed within a session.
- Revisioned universe model defaults are available through the API, CLI, and Platform. Existing sessions retain their selected model when defaults change.
- Standalone transcription supports configured OpenAI and compatible endpoints, start/read/cancel operations, and channel voice preparation. Browser dictation produces an editable draft; the composer also supports attachments and model selection.

**Context reliability and workspace tools**

- Automatic compaction and bounded recovery from typed context-length failures help long-running sessions continue. Manual **Compact context** actions in session and bot menus show queued, in-progress, and failed requests. Provider-native compaction is used where supported, with route-aware summarization fallbacks.
- Authorized callers can replace or redact active user messages and tool results while a run is inactive. Repair changes future model input; original events and blobs remain retained.
- Stable context catalogs avoid duplicate prompt text during metadata-only refreshes. Provider-safe media normalization and request budgets preserve original files and handles, with clearer provider rejection reporting and Chat Completions replay/tool fixes.
- Workspaces gain uploads, downloads, file/folder operations, PDF and blob previews, and persistent VFS file references in transcripts. Session resource summaries, environment activation, reply copying, and universe appearance are also improved.

Thanks to @cptspacemanspiff for the CLI chat improvements in [#101](https://github.com/smartcomputer-ai/lightspeed/pull/101).

[Full changelog: v0.3.0...v0.4.0](https://github.com/smartcomputer-ai/lightspeed/compare/v0.3.0...v0.4.0)
Loading