Release notes

ExoHuman Cloud instances are moved to new releases by our team in planned windows — see Maintenance and updates. Your instance’s version is shown in Settings → Instance → Version and maintenance.

v1.4.0 — current release

Cheaper routines, agents that can change runtime, and departments you can share.

• Per-schedule pre-check hooks. An agent can ship a pre-check per routine at ~/.exohuman/pre-check.d/<schedule-name>. A hook can answer with a JSON decision: {"fire": false, "reason": "…"} skips the run with no model call, {"fire": true, "context": "…"} adds context to the schedule message, and {"fire": true, "message": "…"} replaces it. Hooks receive EXOHUMAN_SCHEDULE_NAME and EXOHUMAN_SCHEDULE_ID.
• Switch an agent’s runtime. Settings → General → Runtime (or PUT /api/agents/{name}/runtime) moves an agent between runtimes keeping its workspace, credentials volume, schedules and run state; its instruction file is carried over to the new runtime’s file.
• Portable department recipes. A department can be exported with each agent’s runtime, model, skills, required credential names and instructions — never credential values, memory or chat history — and deployed elsewhere.
• Creation-time instructions reach the agent, seeded once into its instruction file.
• Clearer error when shared file storage is not writable.

Runtime switching does not convert credentials: the new runtime needs its own sign-in or provider key.

v1.3.x

Two maintenance releases (v1.3.1, v1.3.2) fixing release tooling; no product changes.

v1.3.0

• Every agent surface works on every worker. Agents scheduled on worker machines get full Files, skills, settings, stats and execution dispatch. The interactive terminal and live streaming stay on the control plane and say so clearly.
• Self-healing workers. When a lost worker rejoins, its agents’ circuit breakers reset automatically.
• Pi is the default runtime, with any connected model; each person picks their own default runtime in Settings → My account.
• Mobile top bar with the same primary navigation as desktop.
• Workspace Projects get their full backend: members, tasks, log, files and steward.

v1.2.0

• Settings → Models is the single model source. Administrators connect providers and enable models; each is selectable everywhere as <connection>/<model>.
• Every runtime routes connected models the same way — Claude Code, Codex, Gemini CLI, OpenCode and Pi.
• Hot model updates. A newly enabled model is usable on the next execution without restarts.
• Actionable errors when a model can’t be resolved.
• Safer updates: concurrent updates are refused and every update is backed up and can roll back.
• Foundation for distributed execution across worker machines.

v1.1.0

• Release channel. Instances run pinned, tested releases and are updated with backup, verification and automatic rollback.
• Per-user feature access — administrators enable optional features for specific people.
• Private by default. Nothing — telemetry, bug reports, docs Q&A — contacts external services unless explicitly configured.

v1.0.0

Released 2026-10-20. Built around one idea: an agent takes a seat on the team.

• Agents are assigned to people — whom they serve, who approves for them, who collaborates. Workspace shows each person a role card with objectives.
• One Inbox with platform-native approvals across every agent.
• Gated skills that require approval by a role before they run.
• Self-change grants — agents change their own skills or schedules only when granted.
• Projects, business metrics and objectives, suggestions and readiness checks.
• One Settings home per agent and a restored agent header.
• Git sync that pulls — auto-syncing agents fetch and rebase on a heartbeat.
• Workspace-only users — the user role signs in to the Workspace only.
• A broad security sweep: agent keys can no longer reach beyond their grants.

v0.9.5

Released 2026-09-17.

• Workspace becomes where the work happens: every agent gets a canvas; a rail with Info, Loops, Files and Canvas; resizable columns.
• Several topic-scoped chats per agent with a pinned Main chat.
• Running work shows in the chat as a live execution card.
• Agents can ask a named person a question and be woken by the answer.

v0.9.0

Released 2026-08-17.

• Workspace — the client surface — available in every instance: chats with one or several agents, @mention to bring an agent in, / playbooks, streaming turns.
• Per-client rate limits, sign-out and re-entry blocking.

v0.8.5

Released 2026-07-26.

• Skills library with versioned packages and a managed skill runner.
• Rooms — shared multi-agent sessions with budgets.
• Agent display names separate from the immutable id.
• Local, private product insights.

v0.8.0

Released 2026-07-08.

• Two-factor authentication (TOTP).
• Configurable parallel-task capacity per agent.
• Expose an agent via MCP as its own tool.
• Per-agent model override for public channels.
• Git status, sync, log and pull as direct MCP tools; agent reports.
• Idempotency keys for outbound side effects.

v0.7.0

Released 2026-06-23.

• PostgreSQL as the production database, with managed migrations.
• Codex runtime alongside Claude Code.
• Execution-correctness work and credential hot-reload rotation.

v0.6.x

• Unified executions dashboard and per-schedule analytics.
• Loops — bounded sequential runs with stop conditions — with a web UI.
• Phone calls over Gemini Live.
• Agent overview dashboard; Guardrails renamed to Settings.
• Pre-check hooks for schedules, audit retention, soft-delete and data retention policies, default chat timeout raised to 60 minutes.