LanternDownload

Lantern release notes

What changed in each version.

v0.8.2Latest

Reliable relay presence

Pairing stays available without spending the relay database’s command budget on constant presence writes.

  • Throttled Redis presence refreshes independently of client heartbeat traffic.
  • Extended presence leases while retaining connection-close cleanup.
  • Started heartbeats only after registration succeeds.
  • Preserved reconnect backoff when relay registration fails.
v0.8.1

Provider-style concurrency

Lantern now admits independent provider turns without treating every request as a retained conversation.

  • Removed the two-session ceiling from request-scoped AI SDK and compatibility calls.
  • Added optional per-provider turn limits with bounded, abort-aware queues.
  • Shared one supervised Codex App Server process across independent logical threads.
  • Added relay acceptance signaling for requests waiting on provider admission.
v0.8.0

Native tools everywhere

Apple Intelligence, Codex, and Claude can now call app-defined tools through their primary native interfaces.

  • Replaced prompt-emulated tool JSON with structured provider-native calls.
  • Added Apple Foundation Models tools, Codex dynamic tools, and a scoped Claude MCP bridge.
  • Kept provider turns alive while OpenAI, Anthropic, and AI SDK clients execute tools.
  • Added tool capabilities and direct WebSocket tool calls and results.
v0.7.2

A public update channel

Lantern’s signed Mac updates now ship through stable public download paths.

  • Moved the signed appcast and Mac archives to durable public Blob paths.
  • Made release publication upload archives before making the appcast visible.
  • Kept private GitHub Release copies of every Mac artifact for provenance.
  • Aligned clean site installs with the current Lantern workspace package.
v0.7.1

Automatic Mac updates

The Lantern menu-bar app can now keep itself current through signed background updates.

  • Added automatic update checks and background installation with Sparkle.
  • Added a manual Check for Updates action to the app’s settings menu.
  • Embedded the updater in the universal Apple silicon and Intel app bundle.
  • Made tagged releases publish an Ed25519-signed update feed and archive.
v0.7.0

Apple Intelligence on device

Approved apps can now use Apple’s system language model on a supported Mac.

  • Added an Apple Foundation Models provider with streaming and multi-turn sessions.
  • Bundled the native arm64 model helper inside the existing Swift menu-bar app.
  • Made setup offer Apple Intelligence only when the system model is ready.
  • Kept Codex and Claude as signed-in online options alongside the on-device model.
v0.6.4

A clearer connection

Apps now share one lit and unlit connection interface that shows whether the paired Lantern is actually reachable.

  • Added live relay-presence checks without making an inference request.
  • Separated saved browser pairing from connected and unreachable states.
  • Added button, dialog, and inline panel treatments to LanternConnect.
  • Added the custom unlit 3D lamp and refreshed the demo and integration docs.
v0.6.3

Tools and partner badges

Apps can call client-defined functions through Lantern and show visitors that the integration is available.

  • Added OpenAI-compatible client-defined tool calling for streamed and non-streamed responses.
  • Preserved assistant tool calls and tool results across multi-turn tool loops.
  • Supported automatic, required, disabled, and named tool choices.
  • Added responsive Works with Lantern badge and marketing section components for React sites.
v0.6.2

Lantern in the menu bar

Lantern now has a native Mac companion for setup, status, and controls.

  • Added the native Lantern menu bar app.
  • Showed installed CLIs, Lantern's online state, and approved apps in one place.
  • Added start, stop, restart, setup, logs, and login controls.
  • Published a universal Mac download for Apple silicon and Intel.
v0.6.1

Clearer update messages

Apps can now explain when Lantern needs an update, while release checks keep the site and package in step.

  • Added an explicit version to the pairing protocol.
  • Showed an actionable update message for pre-tunnel Lanterns.
  • Made the package manifest the runtime version source of truth.
  • Blocked package drift under an already-published version.
v0.6.0

Connect from a web app

A web app can now use the Lantern each person approved.

  • Added a persistent local tunnel identity and outbound WebSocket connection.
  • Added the Vercel WebSocket relay with Redis routing across function instances.
  • Stored the authenticated tunnel binding in the app's signed HttpOnly pairing session.
  • Added lantern.server(request) for a Vercel AI SDK provider bound to the connected user.
  • Made provider IDs the canonical model names instead of maintaining a parallel alias map.
v0.5.0

Agent runs in the AI SDK

Trusted local tools can ask for one workspace run through the same AI SDK.

  • Added providerOptions.lantern.agent with one-request working directory, access, and model overrides.
  • Preserved cache, model, cost, terminal, and token telemetry through AI SDK provider metadata.
  • Made Codex and Claude agent sessions ephemeral and moved Relay onto generateText.
v0.4.0

Local workspace runs

Desktop tools can ask for one local workspace run, with read or write access chosen each time.

  • Added runLocalLanternAgent through @sunbeam-za/lantern/node for trusted desktop tools.
  • Added the loopback-only /v1/agent/runs endpoint with per-request read or write access.
  • Removed the hosted relay path so web app requests and workspace access have separate security boundaries.
v0.3.0

Connect Lantern for web apps

Lantern added a user-approved connection from web apps to Codex or Claude on each user's computer.

  • Added LanternProvider and LanternConnect for fixed-address browser pairing with explicit origin approval.
  • Added Vercel AI SDK providers, chat handlers, and direct browser transports.
  • Added OpenAI-compatible image generation and editing backed by Codex.
  • Published dedicated react, client, and ai-sdk package exports.
v0.2.0

Fresh sessions by default

The public API moved away from saved projects and made fresh, temporary sessions the default.

  • Mapped models directly to providers and removed project-based public configuration.
  • Made fresh temporary directories the default and disabled Claude tools by default.
  • Improved provider discovery, including Codex bundled with the macOS ChatGPT app.
  • Simplified setup, examples, and documentation around fresh temporary sessions.
v0.1.0

Initial release

The first release added a local service for sending requests through signed-in coding CLIs.

  • Introduced the local gateway with Codex and Claude CLI providers.
  • Added OpenAI- and Anthropic-compatible streaming HTTP APIs.
  • Added authenticated WebSocket sessions with start, continue, cancel, and close controls.
  • Shipped the first documentation site, configuration schema, and standalone client example.