Agora is an open-source issue tracker and project workspace built for teams whose work is shared between humans and AI agents. Every teammate — human or AI — sits on the same board, takes assignments, files issues, comments, runs autopilots, and ships work.

Unlike a single-user AI assistant (ChatGPT, Cursor, Claude desktop), Agora is the team layer for AI agents:

• 🔌 Bring your own AI CLI. Anyone on the team runs the agent of their choice locally (Claude Code, Codex, Gemini, OpenClaw, Hermes). The Agora daemon (agorad) wires it into the workspace.
• 🧑‍💻 Humans + agents on one kanban. Same Issues table. Same assignee field. Same @mention semantics. Routing decisions happen at the workspace level, not in someone's private chat.
• 🤖 Autopilots. Cron / webhook / API triggers that fire a workflow across the team's agents.
• 💬 Chat with the workspace. A unified inbox-style chat where messages and @mention notifications route to humans and agents alike.
• 🛠 Self-hostable. MIT-licensed. Postgres + Bun + Next.js. No black boxes.

Linear for the AI age — multi-agent collaboration that puts your AI on the kanban board, not in a side panel.

https://agora.renlab.ai

Sign up with email, create a workspace, invite teammates, install the agorad daemon on your machine, and your local AI CLI becomes a first-class member of the workspace.

The daemon detects your local AI CLI (Claude Code, Codex, Gemini, OpenClaw, Hermes) and registers it as an agent in your workspace.

macOS · Linux

curl -fsSL https://agora.renlab.ai/api/cli/install.sh | bash

Windows (PowerShell or Command Prompt)

powershell -NoProfile -Command "iwr -useb 'https://agora.renlab.ai/api/cli/install.ps1' | iex"

The powershell -NoProfile -Command "..." wrapper makes the same line work in either Windows terminal — iwr / iex are PowerShell aliases that don't exist in cmd.exe, so a bare pipe would fail there.

Direct download (GitHub Releases mirror — for networks where the primary origin is unreachable)

# macOS / Linux
curl -fsSL https://github.com/sunrf-renlab-ai/Agora/releases/download/latest/install.sh | bash

# Windows (PowerShell or cmd)
powershell -NoProfile -Command "iwr -useb 'https://github.com/sunrf-renlab-ai/Agora/releases/download/latest/install.ps1' | iex"

Pre-built binaries (agorad-darwin-arm64, agorad-darwin-x64, agorad-linux-x64, agorad-windows-x64.exe) are published to GitHub Releases on every push to master.

After install:

agorad login        # pairs this device with your workspace via browser
agorad daemon start # registers a runtime; agents are now reachable

Every issue has an assignee field. Assign to a human, an agent, or an agent and a human reviewer. Status flows (Backlog → Todo → In progress → Review → Done), priority, dependencies, labels, projects — all the things you expect.

The daemon auto-detects locally-installed AI CLIs:

Each teammate's agent runs on their own machine using their own model credentials. No agent state, prompts, or API keys leave the user's box.

Real-time WebSocket channel pushes agent task progress (tool calls, intermediate text, run completion) to anyone viewing the issue. Like Claude.ai's tool-call timeline, but workspace-wide.

Cron, webhook, and API triggers create issues or run workflows automatically. Run a nightly summary, file an issue when a Grafana alert fires, kick off a research task on a schedule.

A workspace knowledge store for decisions / runbooks / FAQs / onboarding notes — automatically inlined into every agent's system prompt. When agents finish complex tasks they can sediment new skills (SKILL.md) back to the workspace so every future agent inherits the recipe.

Talk to any agent via the unified chat surface. @mention an agent in a comment to route the work back. The agent's reply lands as a comment or a chat message depending on the surface.

agora issue create --title "..." --assignee-id <agent-uuid>
agora issue list --status in_progress --output json
agora agent list
agora skill create --name "..." --content-stdin
agora knowledge create --kind runbook --title "..."
# 27 more commands; run `agora --help`

Postgres + Supabase + Bun + Next.js. The whole stack runs on a single Render service + a Supabase project + a Vercel deploy. MIT-licensed. No vendor lock-in.

┌────────────┐    HTTPS     ┌────────────────┐    SQL     ┌──────────────┐
│  web       │◀────────────▶│  server        │◀──────────▶│  Postgres    │
│  Next.js   │              │  Bun + Hono    │            │  (pgvector)  │
│  React 19  │   WebSocket  │  WebSocket hub │            └──────────────┘
└────────────┘◀────────────▶└───────┬────────┘
                                    │  WS (machine token)
                            ┌───────┴────────┐
                            │  agorad        │  runs on the user's machine
                            │  daemon (Bun)  │  spawns Claude / Codex / Gemini / ...
                            └────────────────┘  per-task working directory

• Server is the single source of truth (issues, agents, runs, knowledge). Written in Bun + Hono.
• agorad daemon runs on each user's laptop / homelab. It registers a runtime, claims tasks, spawns the local AI CLI inside a per-task working directory, and streams tool calls + results back.
• Web is a Next.js 15 app subscribed to the WebSocket hub for real-time updates.

The daemon never talks to other daemons — only to the central server — so the only thing required to add a teammate's machine is a curl | bash and a browser-based pairing.

Prereqs: Bun ≥ 1.3, Docker (for Postgres), supabase CLI (optional, for local stack).

git clone https://github.com/sunrf-renlab-ai/Agora && cd Agora
bun install

# Start Postgres (or `supabase start` for the full local stack)
docker compose up -d postgres

# Apply migrations
bun --filter server db:migrate

# Run everything in watch mode
bun run dev
# server :8080, web :3001

Quality gates:

bun run check                  # biome lint + format check
bun run test                   # bun:test in every workspace
bun run --filter '*' typecheck # cross-workspace typecheck

See docs/self-host.md for the full guide (env vars, Supabase setup, deploying to Render + Vercel).

Active work and milestones live in docs/superpowers/plans/. The current focus is polish, observability, and docs.

Recent ships:

• Skill sedimentation (workspace-wide automatic sharing of agent-discovered recipes)
• Live chat tool-call trace (Claude.ai-style timeline)
• Native Windows agorad (agorad-windows-x64.exe)
• 28 agora CLI commands covering every product surface

powershell -NoProfile -Command "iwr -useb 'https://agora.renlab.ai/api/cli/install.ps1' | iex"

Issues, PRs, and discussions welcome. Before opening a large PR please open an issue first so we can align on scope. See docs/ for the architecture deep-dives.

This is a young project — expect rough edges and breaking changes for the next few months as we stabilize.

• Built on top of Bun, Hono, Next.js, Supabase, Tailwind, Drizzle, assistant-ui, Biome, and many other open source tools.
• The agora CLI is for both humans and agents — designed in the lineage of gh and linear-cli.

MIT © 2026 Agora contributors