Skip to content

Releases: caelis-labs/caelis-bot

v0.21.0

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 30 Oct 09:46
376400a

0.21.0 (2026-10-30)

⚠ BREAKING CHANGES

  • Bundle official Caelis Core v0.75.0 (fe8fdbd056ace719d7b3f302d8e78151c3764e97) with protocol files from the same source. Bot requires runtime-agent-team-v1 and rejects an incompatible Host, including a busy older Host awaiting idle replacement. Upgrade compatible Bot and Runtime together: older Bots can reject the new Runtime's reviewer state. Historical reviewer records remain evidence, not configuration authority. (#263)

Features

  • Use the Runtime's global Agent Team for Guardian and internal helpers while keeping the primary Bot and Worker models independently selectable. Preserve existing sessions, work and receipts. (#263)
  • Offer low-frequency, optional memory-review invitations after accumulated work and recent repeated compaction. The Bot may defer; an invitation does not force Dream, delete context or diagnose an unhealthy session. Missing metrics remain unavailable. (#261)

Bug fixes

  • Unify Worker reports, approvals, questions, reminders and callback delivery through the resident system-input path. Preserve original event identities and accepted/rejected/unknown receipts; internal events do not appear as user messages or grant new user authorization. (#258)
  • Make Dream handoff available before the new session accepts its first input. Preserve user corrections and Stop priority, reconcile original renewal receipts before another Dream, and retain terminal-result fences across reconnects without replaying accepted inputs. Workers can continue running during handoff. (#262)
  • Keep the capture selection panel at the lower right and send only the selected screenshot, without an additional full-screen image. (#257)

Maintenance and verification

  • Refine the personal-assistant skill and its task-specific references. Distinguish Notebook core memory and notes from Runtime memory, and retain original business records as task and approval authority. (#254)
  • Preserve sanitized storage-error causes and native failure evidence; verify cancellation, original receipts and bounded storage recovery independently. These changes do not establish the root cause of every historical HTTP 500 or Windows storage timeout. (#260, #263, #262)
  • Memory review and Dream continuity have deterministic regression and real-runtime synthetic-protocol coverage. Autonomous model decisions and the semantic quality of retained or reorganized memory have not been independently validated.

v0.20.0

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 28 Oct 21:45
9d7fc78

0.20.0 (2026-10-29)

Features

  • Bundle official Caelis Core v0.74.0, with matching public protocol provenance. Add dynamic asynchronous and idle compaction settings, retire the old MCP cache toggle, and preserve cache read/write usage presence and TTL. Background compaction jobs remain separate from foreground turns. (#252)

Bug fixes

  • Keep Worker completion feedback and subsequent delegation in the normal Agent Loop. Preserve original user/tool identities, recover stale Worker projections through native reads, and retain accurate pre-dispatch rejection evidence. Completed Workers remain available for explicit continuation. (#250, #249)
  • Preserve Worker questions, approval targets and answer receipts across continued execution generations; reject stale finish callbacks without closing newer work. (#247)
  • Preserve actual Worker rejection reasons and distinguish authorization failures from retryable admission outcomes. (#246, #248)
  • Align dark backgrounds, keep completed Worker history accessible without crowding current task choices, and smooth split-view output while preserving corrections, final output, task switching and reading position. (#251, #245)
  • Recover macOS shortcuts after conflicts and transient SSH disconnections without losing existing work. (#243, #240)

Maintenance

  • Update Wails to v3.0.0-beta.34, SQLite to v1.61.0 and Vite to 8.3.5. (#239)
  • Correct persistence and plugin lifecycle fixtures exposed on Fedora and preserve complete Windows test coverage in bounded partitions. These test changes do not enable native OAuth on Linux. (#242)

v0.19.2

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 27 Oct 14:48
a099230

0.19.2 (2026-10-27)

Bug fixes

  • Batch shared Weixin replies by urgency and refresh delivery windows on incoming messages. Keep previews distinct from full-text delivery, and let /r resume retained results under their original task and delivery receipts. (#235)
  • Permanently delete local connections while offline, including their legacy records, retained connection proofs and connection-only credentials. Migrate previously archived entries through the same cleanup while retaining task history with its environment unavailable; incoming authorization remains independent. (#236)

Bundled runtime

  • Update Caelis Core to v0.73.3, including child-message outcome fixes and the refreshed native model catalog with GPT-6.1 Terra. Keep the public OpenAPI and consumed wire protocol unchanged. (#238)

v0.19.1

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 27 Oct 03:14
0d98742

0.19.1 (2026-10-27)

Remote connections and SSH

  • Simplify SSH login with an existing OpenSSH Host or manual host, port and username using a password or local key. Remove SSH file-management entry points and bot-core file-editing guidance while keeping legacy private files compatible internally. (#233)
  • Restore local editing and removal for legacy connections. Removal archives the list entry while retaining task routes and proofs; editing does not reconnect, and incoming authorization remains separate.
  • Recover provably undelivered operations under their original identity and preserve rejection receipts across restart, without replaying late requests.
  • Explicitly apply saved candidate addresses for pending or uncertain connections through the original pairing receipt. Consume an accepted candidate atomically while preserving newer saved addresses or names.
  • Reclaim idle archived SSH channels while protecting task, shared and active pairing references.
  • Reject invalid address and SSH ports before normalization or submission.

Validation scope: the source PR passed all five final CI checks and real local isolated OpenSSH / controlled Daily peer operations. External Fedora Bot/Runtime acceptance is not claimed. Windows native SSH remains outside the supported transport matrix; new key selection is desktop-only and passwords/passphrases are not retained across sessions.

v0.19.0

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 26 Oct 13:41
3e68d73

0.19.0 (2026-10-26)

Remote machines and chat

  • Simplify remote machines into one connection and authorization list with clear direction, editing and SSH controls. Keep readable machine names consistent across dispatch, task lists and observation windows without changing connection identity or task routing. (#231)
  • Unify desktop remote and Web Worker chat, preserve reading position and unsent Web drafts, and open local tasks in their original terminal session. Questions and approvals show the selected answer and actual processing result while preserving original requests, receipts and uncertain submissions. (#231)

Worker outcomes

  • Distinguish confirmed local or native task-creation refusals from uncertain dispatch, preserving attachment refusal handling and the original request. (#232)
  • Deliver bounded Worker failure details through task reads, observation, remote transport and the original host report. Use validated native failure facts where available; retain unknown when the runtime cannot establish a cause. (#232)
  • Suppress a redundant interrupted-task notification only after a confirmed manual stop of that exact native target. Unconfirmed stops, completion or failure races, and later turns still report their outcomes. Preserve original-task continuation guidance for occasional Codex capacity failures without adding model switching or automatic retry loops. (#232)

Task interaction lifecycle

  • Let the Bot finish inspected terminal or idle task interactions after validating the original owner and result. Bound temporary questions, approval payloads, drafts and feedback while keeping minimal uncertain receipts tied to their original operations. Finishing does not answer questions, approve actions, stop native work or replay uncertain effects. Preserve existing callbacks and prevent collected history from resurfacing as new questions. (#229)

v0.18.3

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 25 Oct 18:42
52e5d7a

0.18.3 (2026-10-26)

Remote connections

  • Connect to another running Bot by address or SSH Host with asynchronous approval on the target. The initiating Bot keeps task output, history, files, questions and approvals; the target retains its own accounts, models and isolated execution. Each direction requires separate authorization. SSH supplies a network channel only; it does not install or configure the remote Bot. SSH convenience is available on macOS/Linux; Windows supports direct Bot connections. (#221)

Bug Fixes

  • Preserve original Worker prompts and deliver host-validated answers to the original Worker even while the main Bot is idle. Bound question and pairing retention and large tool reads while preserving history, callback compatibility, original request identities and uncertain receipts without replay. (#225)

Assistant guidance

  • Refine bot-core around a continuing personal assistant: follow the user's goal and latest corrections, arrange future work, track Workers, verify results and finish delivery. Add guidance for reusing and reclaiming workspaces, test data, caches and intermediate artifacts according to ownership and active dependencies. This updates packaged guidance; it does not add automatic garbage collection. (#227)

Bundled dependencies

  • Upgrade the bundled Desktop World Plugin to v0.1.3 with CUA 0.38.0. Include browser discovery/next_call and idle context-reset guidance while retaining the single exec interface, existing authorization and original-receipt recovery. No global Desktop World installation is required. (#228)

v0.18.2

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 22 Oct 12:18
28d2394

0.18.2 (2026-10-22)

Bug Fixes

  • chat: preserve IME input, model menu layering, and completed reply display (d4912bd)
  • telegram: preserve complete Web pairing links and bound Start guidance (24266f3)

v0.18.1

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 21 Oct 09:13
1acfd0b

0.18.1 (2026-10-21)

Bug Fixes

  • install: identify Linux bootstrap download requests (#215) (cbd8867)

v0.18.0

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 21 Oct 08:34
46fcfaa

0.18.0 (2026-10-21)

Features

  • add shared headless Web surface and managed Linux distribution (#208) (c2117bd)
  • expose explicit shared Runtime experiment controls (#212) (e361726)

Bug Fixes

  • bundle the verified Desktop World v0.1.2 plugin (#211) (92ea578)
  • refine shared chat, setup, tasks and Web sessions (#213) (59db654)

v0.17.1

Choose a tag to compare

@caelis-character-publisher caelis-character-publisher released this 19 Oct 14:05
2d94f10

0.17.1 (2026-10-19)

Bug Fixes

  • schedule: allow user-arranged follow-ups while the desktop is locked or its presence is unknown, and fix macOS presence decoding. Desktop-input tools retain their own permissions. (#202)
  • schedule: retire the separate care tools and dispatcher; migrate existing arrangements into the shared scheduler while preserving original authorizations, pending work and uncertain receipts without replay.
  • schedule: prevent retained native IDs from corrupting the event ledger after extended app exit, preserve shared grants when removing historical arrangements, and enforce consistent capacity limits for unfinished work.
  • schedule: restore lookup by original native ID and name, defer busy submissions under their original IDs, and preserve original-channel attribution for scheduled follow-ups.