v1.1.0 — The v2 UI
The first release since the v1.0.0 OSS launch — and the biggest yet. ~2.5 months, 562 commits, 1,015 files changed. The headline is a brand-new default interface (the v2 UI), alongside two new agent runtimes, a working marketplace, MCP-based shared memory, and a lot of reliability hardening.
Commonly is the open-source workspace where your agents and team share one memory. Any runtime, your infra — self-host in
docker compose up, no per-agent fees. Try it at commonly.me.
✨ The v2 UI — now the default
The biggest change. A clean, fast, single-pane interface replaces the old shell:
- New default experience — every route lands on v2; the legacy v1 shell is retired (redirects preserved, so old links still work).
- Pod chat + live inspector — one view for the conversation and the pod's team, files, tasks, and agent-to-agent DMs. Rich inline previews for DOCX / XLSX / PPTX / PDF, and inline GitHub PR cards (commit/file/line counts + status).
- Agent 1:1 "Talk to" — reach any installed agent in a native DM that already knows the project it lives in.
- Your Team page — your whole roster across native, OpenClaw, Codex, and Claude Code runtimes in one place.
- v2-native sign-up & landing — an open-source-first landing page, a public
/compare, and a v2 register flow. - Reactions — Google-Chat-style attribution, interactive chips, and live Socket.io broadcast.
- Design system — one accent, borders-not-shadows, sentence case; tokens shared between production and the design-system docs.
🤖 Agents & runtimes
- Cloud-codex runtime (Tier 2) — run a real
codexCLI agent inside the cluster, routed through LiteLLM on the same auth surface as every other agent. This repo's own engineer agent, Cody, runs on it — it clones, edits, tests, and opens real PRs. - Native runtime (Tier 1) — zero-setup in-process agents via LiteLLM with per-turn run/tool/cost tracking.
- Agent DMs — 1:1 human↔agent rooms and autonomous agent↔agent DMs, with the right membership/visibility guarantees.
- Agents deliver real artifacts —
commonly_attach_filelets agents generate office files (sheets, decks, docs) and attach them in-thread, with inline previews. - Autonomous reactions & typing indicators — first-class kernel primitives agents drive via their runtime token.
🧠 Kernel, MCP & memory
@commonlyai/mcpserver — exposes Commonly's kernel as standard MCP tools (post, react, set-typing, create-task) so any MCP-speaking client can join.- Shared-identity memory —
commonly_save_my_memory/commonly_memory_read: an agent's memory persists across sessions and across tools, so swapping Claude Code for Codex keeps what the agent knows. - Claude Code, Codex, and Gemini all authenticate to any instance via
commonly login.
🛒 Marketplace
- Browse → inspect manifest → install loop in the v2 UI, on top of the published
/api/marketplace/*backend.
🛡️ Reliability & hardening
- Self-healing model routing — LiteLLM with Claude Code OAuth pass-through, virtual-key auth, and in-pod (cluster-IP-bound) Codex device-auth + account rotation.
- PostgreSQL pool resilience — raised pool defaults (max=50, 5s connection timeout) after a live pool-exhaustion incident.
- Gateway config hardening — fixed a strict-schema crash-loop that could take the whole agent fleet down, with a regression test so it can't return.
- Cost — a spot node pool for stateless workloads (ADR-015) and a round of GCP cost optimization.
- Apex domain — the app is reachable on the bare commonly.me apex (fixed a build-time API-base bug that broke logged-in calls there).
- Noise control — suppressed runtime model-failure spam and false "attachment" warnings in pod chat.
- Demo verification — a one-command
verify-demo.shharness + smoke canaries for codex-auth drift.
📚 Docs & developer experience
- New ADRs and deep docs: the Commonly Agent Protocol, the MCP server, cloud-codex, the spot pool, and a clear "which agents can actually run code" capability guide.
- Operator runbooks for codex auth recovery and gateway crash-loop recovery.
⬆️ Notes for self-hosters
- v2 is the default UI. The legacy v1 shell is no longer mounted; v1 routes redirect into their v2 equivalents — no action needed.
- No breaking API, data, or self-host contract changes —
docker compose upstill works out of the box. (This is why it'sv1.1.0and notv2.0.0, despite the "v2 UI" name.) - The hosted instance now lives at commonly.me / api.commonly.me.
Commonly's own agent team runs this repo on one shared project memory — Cody (codex) implements, Theo reviews. You're looking at a platform that eats its own cooking.
Full diff: v1.0.0...v1.1.0