GitHub Releases

Changelog that follows the repo.

Release notes are pulled from GitHub Releases, then grouped by their Markdown sections so the public changelog stays aligned with what ships.

Subscribe or inspect the source.
Subscribe via RSSBrowse GitHub releases

v0.1.44-beta.15

GitHub

v0.1.44-beta.15 Fixed Restore compatibility with databases created by beta.11 and later by bundling the previously released TODO target-lock migration with its original identity. Tests Verify the beta.11 migration hash and confirm compatible databases are inspected without changing their bytes

Highlights

  • Restore compatibility with databases created by beta.11 and later by bundling
  • Verify the beta.11 migration hash and confirm compatible databases are inspected

v0.1.44-beta.14

GitHub

v0.1.44-beta.14 Added Run automatic memory learning through a persisted, bounded background worker so completed chat turns no longer wait for the memory provider; expose extraction outcomes, Dream maintenance, write policy, and Markdown sharing controls. Add recoverable memory maintenance and revision-aware Markdown import/export with explicit opt-in scopes, validation, and visible conflicts

Highlights

  • Run automatic memory learning through a persisted, bounded background worker so
  • Add recoverable memory maintenance and revision-aware Markdown import/export
  • Keep GPT-6 Astra compatible with the current Codex Responses Lite protocol and
  • Make the Minimal prompt preset truly compact and retain the final Arcade/Snake
  • Preserve queued prompt delivery and terminal cancellation while compaction,

v0.1.44-beta.13

GitHub

v0.1.44-beta.13 Added Let tell forward current, selected prior, and explicit local attachments to another session, then return files produced by that session to the sender. Fixed Restrict TODO target-lock validation to transfer, flash, and verify commands so ordinary shell commands do not inherit stale device-operation metadata

Highlights

  • Let tell forward current, selected prior, and explicit local attachments to
  • Restrict TODO target-lock validation to transfer, flash, and verify commands so
  • Clear the queued state after completed assistant turns with an unknown finish
  • Keep peer-message responses correlated through compaction and bound attachment
  • Add focused regressions for target-lock scope, queued-state release, peer

v0.1.44-beta.12

GitHub

Experimental prerelease from 1429a172098918b0820f39981d53575d1651ee1c. Concurrency remains opt-in.

v0.1.44-beta.11

GitHub

v0.1.44-beta.11 Added Add permission-gated, on-demand Computer Use with bounded semantic inspection, macOS interaction highlighting, strict action and duration limits, and a local smoke harness. Add revisioned TODO target locks for transfer, flash, and verify operations, rejecting stale target/artifact/port data and grounding final reports in recorded command output

Highlights

  • Add permission-gated, on-demand Computer Use with bounded semantic inspection,
  • Add revisioned TODO target locks for transfer, flash, and verify operations,
  • Preserve the active session model, provider, and variant when delivering
  • Return peer-session responses after overflow compaction by preserving and
  • Deliver completed peer-session responses immediately instead of leaving them

v0.1.44-beta.10

GitHub

v0.1.44-beta.10 Fixed Sign and verify the final macOS executable after compilation, before creating release archives. Let updated CLI and desktop clients attach to a healthy shared backend while existing clients own its previous runtime, without starting another database writer or failing update readiness solely on the backend version

Highlights

  • Sign and verify the final macOS executable after compilation, before creating
  • Let updated CLI and desktop clients attach to a healthy shared backend while
  • Require harness authorization before custom, plugin, and MCP tool hooks or
  • Ignore legacy persisted always grants after restart and prevent DWS from
  • Update ChatGPT OAuth's Codex compatibility header to 0.154.0 so GPT-6 Astra

v0.1.44-beta.9

GitHub

v0.1.44-beta.9 Added Add passive prompt-cache policy controls with project, session, provider, and exact-model scope resolution, plus stable cache-lineage fingerprints. Add provider-aware cache observations and conservative smart-mode adapters without enabling keepalive or remote-cache deletion

Highlights

  • Add passive prompt-cache policy controls with project, session, provider, and
  • Add provider-aware cache observations and conservative smart-mode adapters
  • Preserve ChatGPT OAuth Responses Lite affinity while rotating it when the
  • Honor explicit cache disablement across OpenAI, Claude Code, and transformed
  • Keep generated cache metadata and lineage deterministic across session and

v0.1.44-beta.8

GitHub

v0.1.44-beta.8 Added Add a bounded Agent View command panel with actionable peer-session history, direct session navigation, and explicit acceptance or rejection controls. Add model-family prompt compatibility, bounded tool discovery, context profiles, and runtime capability reporting without changing provider identity

Highlights

  • Add a bounded Agent View command panel with actionable peer-session history,
  • Add model-family prompt compatibility, bounded tool discovery, context
  • Preserve queued prompts and peer delivery across targeted cancellation,
  • Harden shared-server leases, lock ownership, shutdown cleanup, and targeted
  • Add focused regressions for queue/cancellation recovery, shared-server

v0.1.44-beta.7

GitHub

v0.1.44-beta.7 Added Add passive prompt-cache controls by project, session, provider, and exact model through mendcode cache status|enable|disable and the cache config block, without enabling keepalive or remote-cache deletion. Add conservative smart-mode adapters, cache lineage fingerprints, and provider cache observations that keep unsupported bindings unknown

Highlights

  • Add passive prompt-cache controls by project, session, provider, and exact
  • Add conservative smart-mode adapters, cache lineage fingerprints, and
  • Keep ChatGPT OAuth Responses Lite transport affinity separate from managed
  • Remove generated cache keys and provider annotations when the effective
  • Add configuration, adapter, lineage, provider-transform, and ChatGPT OAuth

v0.1.44-beta.6

GitHub

v0.1.44-beta.6 Fixed Complete terminal loop runs without waiting on the owner session that is executing run_once, preventing a circular wait and duplicate notification. Deliver background loop notifications asynchronously while returning direct run_once results to the calling session

Highlights

  • Complete terminal loop runs without waiting on the owner session that is
  • Deliver background loop notifications asynchronously while returning direct
  • Keep Send now attached to the first visible queued prompt when synthetic
  • Preserve transcript follow mode when a viewport resize or tall Shell output
  • Add native PTY coverage for queued notifications, transcript paging and

v0.1.44-beta.5

GitHub

v0.1.44-beta.5 Fixed Pin the release dependency tree to fixed @ai-sdk/provider-utils 4.0.33, hono 4.13.5 and js-yaml 4.3.2 versions, with matching provider types. Make repeated Esc cancellation settle active, queued, peer, and owner-wake work without replaying a cancelled prompt; a fresh explicit submission is required to resume the session

Highlights

  • Pin the release dependency tree to fixed @ai-sdk/provider-utils 4.0.33,
  • Make repeated Esc cancellation settle active, queued, peer, and owner-wake

v0.1.44-beta.4

GitHub

v0.1.44-beta.4 Added Native/portable compaction checkpoints and explicitly configured compound-model workflow plans, snapshots and execution accounting. AI configuration inspection, planning, validation and permission-gated apply through the shared backend, CLI and native tool

Highlights

  • Native/portable compaction checkpoints and explicitly configured compound-model
  • AI configuration inspection, planning, validation and permission-gated apply
  • Update the transitive query-string parser to qs 6.16.0, fixing
  • Normalize replayed tool arguments and bound connection retry behavior.
  • Fence targeted turn and shell cancellation without escalating a stale target

v0.1.44-beta.3

GitHub

v0.1.44-beta.3 Fixed Update Drizzle ORM and Kit to the fixed 1.0.0-beta.20 versions, including correct SQL identifier escaping (GHSA-gpj5-g38j-94v9). No schema migration is generated

Highlights

  • Update Drizzle ORM and Kit to the fixed 1.0.0-beta.20 versions, including
  • Backend-owned Smart Approval review, exact action grants, history and revocation,
  • mendcode context for estimated context composition, provider-reported cache
  • Stable per-turn memory snapshots and bounded secondary-tool discovery, with a
  • Opt-in Code Mode adapted from the pinned OpenCode v2 interpreter. Nested tools

v0.1.44

GitHub

This stable hotfix carries startup recovery independently of the beta runtime.

Highlights

  • Stop treating a slow database migration as a failed eight-second backend
  • Stop waiting when the owned child exits or reports failure; identify its
  • Reject a recorded, incompatible prerelease schema before opening a database
  • Keep the Windows installer parseable by Windows PowerShell 5.1.
  • Load bundled GitLab and Poe authentication modules through the existing

v0.1.44-beta.2

GitHub

v0.1.44-beta.2 Fixed Let a newly spawned backend finish a large pre-migration database snapshot instead of stopping it after eight seconds. Verified backup/migration phases have a bounded 15-minute preparation window and visible progress messages

Highlights

  • Let a newly spawned backend finish a large pre-migration database snapshot
  • Keep the 30-second client readiness deadline separate from database preparation.
  • Document published beta channels, large-history recovery and stable promotion.

v0.1.44-beta.1

GitHub

This beta introduces release channels and updater recovery. It is an opt-in prerelease; stable installations remain on the stable channel.

Highlights

  • Bound installer downloads, verify checksums, stage replacement atomically,
  • Preserve explicit database paths when starting the shared backend and report
  • Route run and stats through the shared backend. Failed runs return a failure
  • Preserve quotes and backslashes in run arguments and load bundled GitLab/Poe
  • mendcode upgrade channel, channel set stable|beta|nightly, --check, and