Reference · 3 min read
Permission modes
The five steering modes in Miton — Chat, Plan, Confirm, Autoedit, and Full access — and what each allows without asking.
- Product stage
- Pre-release development
- Last reviewed
- 2026-07-23
- Successful result
- The change is visible in the active project.
Who this is for: anyone who wants to control how much the agent can do on its own in the current chat session.
Steering modes (also called execution modes) are a per-session setting. They work together with your workspace trust level — trust sets the outer boundary; mode sets the inner policy. See permissions taxonomy for how the two combine.
For a walkthrough, see Permissions and safety in the handbook.
The five modes at a glance
| Mode | Reads files | Writes files | Runs shell / network |
|---|---|---|---|
| Chat | Yes | No | No |
| Plan | Yes (planning prompts) | No | No |
| Confirm | Yes | Ask first | Ask first |
| Autoedit | Yes | Yes | Ask first |
| Full access | Yes | Yes | Yes (within trust) |
Confirm is the recommended default when you want to stay in control — Miton shows an approval step before file writes, shell commands, and similar actions.
Mode details
Chat
Read-only conversation. Use this when you want answers and exploration without any risk of file or command changes.
Plan
Read-only with planning-oriented prompts injected. Use this to design an approach before switching to Confirm or Autoedit to implement it.
Confirm
Ask before writes and before shell, network, plugin, and memory actions. This is the balanced default for most work — you approve each sensitive step in chat.
Policy summary shown in the UI: Confirm mode requires explicit approval for shell, network, and file writes.
Autoedit
File writes proceed automatically; shell and network actions still require approval. Use this when you trust file edits but want a checkpoint before commands run.
Full access
All capabilities the workspace trust level allows, without per-step prompts. Use only in folders you fully trust. Untrusted or restricted workspaces still block actions regardless of mode.
Changing the mode
| Where | How |
|---|---|
| Mode chip | Click the mode label in the roster bar above the chat input |
| Mode badge | Click the badge above the message list |
| Slash command | Type /mode to list modes or /mode confirm (etc.) to switch |
| Settings | Settings → Advanced → Steering modes sets the default for new sessions |
Mode changes apply on the next tool call. Each session remembers its own mode until you close it.
Per-call overrides
When Miton gates an action, the prompt offers:
- Allow once — proceed for this call only
- Always allow — allow this tool for the rest of the session
- Deny — block the call; the agent continues or stops depending on context
What modes do not do
- Replace workspace trust — an untrusted folder blocks writes even in Full access
- Sync across sessions — each session has its own mode; set a default in Settings
- Apply to other people’s machines — mode is local to your session
See also
- Permissions taxonomy — trust levels and composition rules
- Glossary — Confirm, Chat, and surface names defined
- Slash commands —
/modereference