Features Overview

A map of what Agor does. The first three are the foundation everything else builds on — start there if you’re new. Click into any feature for the full story.

🌳 Worktrees

The unit of work in Agor. A git worktree pinned to a board — branch, isolated environment, sessions, and conversations all attached. Multiple worktrees coexist with their own ports, dependencies, and database state, so parallel work doesn’t collide.

Use when: Always. Worktrees are how Agor organizes everything — you’ll create one for each issue, PR, or experiment.

🧵 Sessions & Trees

Agent conversations with genealogy. A session is one Claude Code / Codex / Gemini conversation pinned to a worktree. Fork to explore alternatives without losing the original path; spawn subsessions for focused subtasks that report back to the parent. The result is a visual tree of how a piece of work actually unfolded.

Use when: You want to branch a conversation, delegate a subtask, or keep a clean trail of what each agent did and why.

🗺️ Boards & Zones

Spatial canvas for worktrees. A board is a 2D Figma-style canvas where worktrees live as cards. Zones are regions on the board that organize work — and can trigger templated prompts when you drop a worktree into them (“analyze this issue”, “review this PR”, “deploy”). Build kanban flows or freeform pipelines.

Use when: You need an at-a-glance view of in-flight work, or want to automate “when something lands here, run this prompt.”

🤖 Assistants

Persistent AI companions that live in your workspace. Long-lived agents with file-based memory, identity, and skills, running on heartbeat schedules and orchestrating work across boards. Inspired by OpenClaw, adapted for teams.

Use when: You want an agent that remembers across sessions and proactively manages a project — not a one-off conversation.

🔌 Agor MCP Server

Anything a user can do in Agor, an agent can do too. The daemon exposes a Model Context Protocol (MCP) server with tools for sessions, worktrees, boards, zones, users, environments, and more. Every session is auto-issued an MCP token, so agents are self-aware and can act on their own context.

Use when: You want agents to operate Agor itself — spawn peers, reorganize the board, create worktrees, prompt other sessions.

This is the foundational layer that makes assistants, scheduled prompts, message gateway, cards, and artifacts all possible.

💬 Rich Chat UX

A CLI can’t show you this. Per-prompt token and dollar accounting, model and effort selectors, live context-window meter, structured tool blocks, todo lists, message queueing, file/user autocomplete, favicon status dots, completion chimes, and Streamdown-rendered markdown with Mermaid + KaTeX + syntax highlighting.

Use when: Always. This is what makes Agor’s surface objectively richer than running a coding agent in a terminal.

👥 Multiplayer & Social

Figma for AI coding. Live cursors, facepile, real-time worktree movement, scoped/spatial comments on boards/zones/worktrees/sessions, attention pulse for completed work, and shared multiplayer terminal sessions via tmux.

Use when: Working with teammates on the same board, or just want spatial awareness of where you’ve been.

🌳 Environments

One-click dev servers per worktree, with no port collisions. Configure start/stop commands once on the repo, with templates that derive unique ports from each worktree’s ID. Everyone on your team gets one-click start/stop on every worktree, with health checks and live logs.

Use when: Working on multiple PRs simultaneously and tired of “kill your dev server so I can test mine.”

⏰ Scheduler

Cron-style triggers for templated prompts. Run a prompt every hour, every Monday, on a custom cadence. Targets a specific worktree or session. Powers assistant heartbeats, daily standups, scheduled audits, periodic backlog grooming.

Use when: Something should happen on a cadence without you remembering to trigger it.

🃏 Cards (Beta)

Generic workflow entities on boards. Not every board entity is a git worktree — support tickets, sales leads, patient records, content pieces. Cards are spatial workflow units that an assistant creates and manages via MCP.

Use when: You want spatial oversight of a non-code workflow, with agents driving state changes.

🎨 Artifacts

Live, interactive applications rendered on the board. Powered by Sandpack, agents scaffold a small app via MCP and the board shows it running — no deploy, no preview URL. Custom dashboards, data visualizations, design mockups, calculators, redline viewers.

Use when: You need visual output during agent collaboration. Agents can build interactive UIs in minutes that would take days to hand-roll.

📨 Message Gateway

Slack and GitHub as portals to Agor sessions. DM a Slack bot or @agor mention on a PR — Agor spins up a session on the right worktree with the right agent, then routes responses back to the platform.

Use when: You want to drive Agor from where conversations already happen — chat, PR threads — without switching contexts.

Reference

• Architecture — System design, data flow, the typed TypeScript client
• SDK Comparison — Capability matrix across Claude / Codex / Gemini / OpenCode
• Security — Deployment modes (simple / insulated / strict), trust boundaries, RBAC