An agent that thinks in notes, not code.
You keep a vault, not a codebase. Copilot v4 takes a capable agent and tunes it for knowledge work. It follows your links, respects your structure, and reads your notes the way Obsidian does, so it feels native to your vault.
Run the best agents in the market. Now they know your vault.
opencode, Claude Code, Codex — the most capable agents available, now working natively inside your vault. Copilot gives each the context, permissions, and Obsidian-native output it needs to do real work; reach for the strongest for the task per conversation. They're the agents you already use, so your subscription comes with you — no second bill, no second key.
Share capabilities with all agents
Add a skill or MCP server once, then switch it on for any agent — opencode, Claude Code, or Codex — with a click. Copilot keeps everything in one place, so there's no setting up three times and no copies to keep in sync.
Your other files stay off-limits.
Copilot limits each agent to your active vault before it starts. It can read and change notes there, and nothing else on your computer. Everything outside is blocked without your permission.
Nothing changes behind your back.
Every write — create, rename, modify, delete — waits as a change you can read before it touches a note. Approve it, reject it, or send it back with a note; nothing is saved until you say yes. Reviewing first is the default, with trusted workflows opt-in to apply on their own.
Your notes never leave your machine.
First party offline support. Point Copilot at a local model — Ollama, LM Studio, or a server you run yourself — and your vault stays put. No note is ever sent to a remote model.
Pick the model that fits the work.
Reach for the right model for the task, deep reasoning, fast drafting, multimodal, or a local model you run yourself. Any model opencode supports, picked per task: your key, your bill, your call.