Releases: cephalopod-ai/cuttlefish
Release list
v0.23.6
Bug Fixes
- Fixed gateway session-query and collaboration routes, plus employee create/editor forms, cron, and settings pages.
- Cron watchdog now interrupts the run's engine when
maxRunMsfires, instead of only clearing the overlap guard and leaving the hung PTY/route promise running indefinitely. js-yaml5.x dropped its CommonJS default export; switched all 16 call sites in this repo to a namespace import soyaml.load/yaml.dumpkeep working (org/employee YAML parsing, config read/write, orchestration config were all silently broken by the default-import pattern).- Claude-interactive's late-Stop-recovery cancellation now logs when it drops a still-armed listener from a previous turn instead of silently discarding it; named the previously-bare mtime staleness tolerance.
Performance
- Slack attachment downloads for a single message now run concurrently instead of sequentially, reducing session-creation latency on high-latency networks.
Security
- Pinned transitive
protobufjs(via@whiskeysockets/baileys>libsignal) to >=7.6.5, closing a moderate denial-of-service advisory (GHSA-j3f2-48v5-ccww) in.protooption parsing.
Dependencies
js-yaml 4.3.0 -> 5.2.2, better-sqlite3 12.x -> 13.0.1, vite 7 -> 8, lucide-react 0.577.0 -> 1.26.0, react/react-dom 19.1 -> 19.2, radix-ui 1.6.2 -> 1.6.7, tailwindcss/postcss patch bumps, @tanstack/react-virtual, imapflow, ws, @playwright/test, eslint, turbo, typescript-eslint, @next/eslint-plugin-next, actions/checkout (CI).
Tests
- Added an end-to-end regression proving the persistent scheduler's multi-writer read-modify-write safety under
BEGIN IMMEDIATEtransactions.
Full Changelog: v0.23.5...v0.23.6
v0.23.5
Corrected follow-up release after the v0.23.4 workflow reached npm without a usable publish credential.
Highlights:
- publishes the corrected
dist/bin/cuttlefish.jsCLI bin path - wires the protected npm-production credential into the provenance-enabled publish job
- updates install and release documentation now that
cuttlefish-cliis public on npm
The published release triggers npm publication first, followed by platform archives and the Homebrew formula update.
v0.23.4
First release with the git tag aligned to the package version, unblocking the npm publish, platform-archive, and Homebrew formula pipelines. Intended as the first published npm release of cuttlefish-cli.
Features
- Collaboration chat routes and UI.
- Orchestration workers data view with end-to-end coverage.
- CLI registers the orchestration command groups.
- Releasable Windows (win32-x64) install path:
scripts/install.ps1, portable zip artifact, and operator docs indocs/INSTALL.md.
Security
- Hardened gateway session, org, and upload handling.
- Fixed auth provider validation and added security regression coverage.
- Bounded security reviewer runs during screening.
js-yamlbumped to 4.3.0.
Bug Fixes
- CLI
limitscommand preserves JSON stdout. - Ollama assistant output stays canonical.
- Checkpoints return coherent decision session state and reject conflicting terminal decisions.
- Gateway accounts completed web session turns.
- Skill installs report idempotent outcomes.
Internal
- Chat sidebar refactored into a project session tree.
better-sqlite3raised to^12.6.2for Node 24 Windows prebuilds.
See CHANGELOG.md for details.
v0.1.0
Cuttlefish v0.1.0
Cuttlefish v0.1.0 is the first public release of a more capable, safer AI-agent gateway: one local daemon for running and coordinating professional coding CLIs such as Claude Code, Codex, Grok, and Antigravity.
This release delivers:
- FleetView, a redesigned operational dashboard with improved navigation, command-center visibility, organization and Kanban workflows, saved data views, and accessibility polish.
- Stronger multi-agent operations: employee hierarchies, manager delegation, approvals, human checkpoints, manager handoff, and opt-in implementer/reviewer execution.
- Durable orchestration foundations: scheduler state, leases, run ledger, worktree support, recovery, and dashboard/API controls.
- More reliable gateway behavior: clear readiness vs. liveness, safer restart/lifecycle handling, custom-home isolation, stronger session dispatch, cron, skills, artifact, and board behavior.
- Security hardening across scoped session credentials, control-plane authorization, connector handling, uploads/files, content screening, SSRF defenses, and auditability.
- Better connector and engine support, including hardened Twilio replay handling, connector reloads, model selection/fallback improvements, and expanded engine validation.
- A comprehensive 225-card playtest library and expanded automated regression coverage.
Cuttlefish remains an early-access release. Advanced orchestration, connector, autonomous-operation, and multi-agent deployments should be run with deliberate configuration and disposable test environments first.
Requires Node.js 24.x.
v0.0.3 - more efficient resource management considerations
What's Changed
- Repair playtest findings: engine-crash surfacing, root sandbox, board 404s, limits refresh, kanban UX by @e3742526 in https://github.com/e3742526/cuttlefish/pull/12
- Resource optimization: OS standby compliance, polling cadence, process lifecycle, memory by @e3742526 in https://github.com/e3742526/cuttlefish/pull/13
Full Changelog: https://github.com/e3742526/cuttlefish/compare/v0.0.2-workinprogress...v0.0.3_resource_mgt_improve
Updated themes and layouts - improved provenance
Major additions:
- Multi-engine orchestration matured substantially. The engine roster now includes Claude, Codex, Grok, Hermes, Antigravity, Pi, Ollama, Kilo, and Aider, with dynamic model discovery, per-session engine/model switching, effort controls, and both chat and PTY/CLI-style execution where supported.
- The web app is now much larger. Beyond chat, it includes Command Center, Talk, Kanban, Orchestration, Cron, Org, Limits, Logs, Skills, Settings, and file/artifact viewing.
- The org model became first-class. Employees are YAML-backed, have departments/ranks/managers, can be created and edited in the UI, can launch direct chats, and now support multi-role execution profiles like mid_pair.
- Kanban evolved into a real work surface. It has department boards, dispatch controls, live session inspection for running tickets, recycle bin/retention, optimistic save protection, manual-only dispatch, and attached resource context.
- Talk became a major feature area. It now has voice-first interaction, persistent conversation flow, delegation graphs, searchable session memory, work rails, breadcrumbed thread navigation, live activity tracking, TTS/STT improvements, and a dedicated multi-agent workspace.
- Matrix orchestration was added. There is now a scheduler, leases, blocked queues, dual-lane execution, worktree-backed runs, artifact capture, retry/recovery flows, queue pause/resume, holds, telemetry, and a dedicated orchestration dashboard plus CLI surfaces.
- Search, history, and transcript handling improved a lot. There is FTS5 message search, structured chat activity rendering, compact tool-call history, better final-history cleanup after turns, and faster chat switching with cache/revalidation.
- Remote and connector workflows expanded. The repo now supports Slack/connector hooks, remote browser pairing/auth flows, inbound IMAP inbox polling and COO auto-ingest, and artifact/file registry routes.
- Reliability and operational hardening increased sharply. Recent releases focused on restart safety, stale-session recovery, queue cleanup, atomic config/job writes, policy cache correctness, auth-cookie security, orchestration lease-release correctness, and broader regression coverage.
- The docs, governance, and release surfaces are much more mature. There is now a maintained README, docs/INDEX.md, architecture/spec/inventory ledgers, cloud-audit baselines, and a much more explicit repo contract.
Full Changelog: https://github.com/e3742526/cuttlefish/compare/v0.0.1-rebrand-jinn...v0.0.2-workinprogress
Cuttlefish v0.0.1
Cuttelfish v0.0.1
Cuttlefish rebrand and runtime separation update (from https://github.com/repo-makeover/jinn which is a fork of https://github.com/hristo2612/jinn). This release is basically (repo-makeover/jinn) with new icons, logo, and everything renamed.
This release completes the application rebrand from Jinn to Cuttlefish across the CLI, daemon, web UI, documentation, templates, runtime paths, and package surfaces.
Highlights:
- Renamed the app to Cuttlefish across public code, docs, templates, UI, package metadata, and CLI surfaces.
- Changed the default runtime home to
~/.cuttlefish. - Changed the default gateway port to
8888so Cuttlefish can run alongside Jinn on the same machine. - Added side-by-side collision guards for runtime home, instance handling, cookies/storage/header naming, and legacy port references.
- Preserved upstream attribution: Cuttlefish is forked from
repo-makeover/jinn, itself a substantially reworked fork of originalhristo2612/jinn. - Added Cuttlefish branding, app icons, favicon, and aquatic/nautical agent avatars.
- Replaced old office-style agent avatars with nautical/aquatic themed avatars.
- Hardened the single-instance model so Cuttlefish does not accidentally allow unlimited local instances.
- Updated setup, status, pair/start docs, templates, and examples to use
cuttlefish,~/.cuttlefish,localhost:8888, andCUTTLEFISH_*. - Improved the Limits page:
- usage cards now show only engines with observed or estimated usage;
- installed CLIs without usage data are grouped separately;
- missing CLIs now show clear “Agent not available” messaging;
- Kiro estimated usage support is wired for configured
kiro-cliinstalls; - the page layout was simplified to avoid awkward card gaps.
- Fixed avatar consistency between Org, Chat, Team, and contact rows so explicit nautical/aquatic org avatars take precedence over stale emoji overrides.
Validation performed includes focused typecheck, lint, unit tests, build, runtime launch, local dashboard checks, side-by-side port checks, residual rebrand scans, and targeted audits for rebrand regressions, failsafe readiness, invariant sync, document stewardship, and UI design consistency.
Known notes:
- Cuttlefish defaults to
http://localhost:8888. - Jinn can remain installed and running separately on its original port.
- Kiro usage is an estimate and requires
kiro-cliplusengines.kiro.creditBudget. - Engines without installed CLIs are now shown as unavailable instead of empty usage cards.
Future direction:
Follow-on code will start to deviate from the philosophy of Jinn. I plan to run both, but for different reasons.
Cuttlefish will more heavily target governance, provenance, and traceability while Jinn will maintain the lightweight system.