Daily Usage
Open Agentic Chat from the ribbon icon or the Agentic Chat: Open chat command.
Send prompts
Type a message and press Enter. Use Shift+Enter for a newline.
The active note is attached automatically unless it is ignored, is a root standing-instruction file (AGENTS.md, CLAUDE.md, or GEMINI.md), or you dismissed the chip for the current session. Standing-instruction files are implicit context whenever present. Add more context with:
@note@note#heading@note^block-id- the folder button in the composer
- drag and drop from the file explorer
- Obsidian context menus for notes, folders, or selected editor text
Watch tool calls
Tool calls render inline in the transcript. You can expand them to inspect arguments, results, and timing.
Mutating calls are controlled by the approval policy. When a call is set to ask, the plugin shows the pending arguments and a line-level diff before anything changes.
Diff previews color removed lines red and added lines green.
Safe, YOLO, and plan modes
Safe honors your configured approval settings.
YOLO auto-approves mutating tools for the current session, except for tools explicitly set to deny.
/plan enters read-only planning mode. It blocks mutations until /endplan restores the previous posture.
Run /compact [instructions] to summarize older turns on demand. The optional instructions are passed to the compaction request so important details survive the summary.
Project workspaces
Use /project to switch between scoped workspaces. Each project can have its own folders, model, style, and tool toggles. /project clear returns to vault-wide mode.
When a project is active, sessions and context are scoped to that project automatically.
Memory
Add long-term memories with /memory add [kind] [scope] <text>. Kinds are preference, fact, instruction, or summary. Scopes are global, vault, or project.
Review stored memories with /memory review. Export them with /memory export. Clear everything with /memory clear --confirm.
The agent retrieves memories explicitly via the search_memory tool; they are never injected automatically.
Semantic indexing
Run /semantic-index start to build a vector index over your current scope (vault, folder, tag, or project). Check status with /semantic-index status and estimate cost with /semantic-index estimate.
Cancel an in-progress build with /semantic-index cancel.
Todo tracking
Use /todo add <milestone> to add a milestone. Update status with /todo set <id> <pending|active|done|blocked>. Track tests with /todo test <id> <not-run|running|passed|failed|skipped>. Attach checkpoint commits with /todo commit <id> <commit>.
The plan tracker panel shows live progress.
Real-time controls
While the agent is streaming a response, you can:
/steer [text]— steer the current turn mid-stream./follow-up [text]— queue a follow-up message behind the active turn./redirect [text]— stop the active turn and answer the new text instead.
Export conversations
Run /export to save the current conversation as a Markdown note in the vault.
Reasoning effort
Override the default thinking level for the next message with /effort [level] (e.g., /effort high). It resets after the next turn.
Undo and retry
Use /undo to revert the last vault mutation made by the agent.
Every assistant answer has copy and retry actions. Click a sent message to rewind to that prompt, edit it, and regenerate from that point.