Skip to content

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 follow the approval policy. When a call is set to ask, the plugin shows the pending arguments and a line-level diff (red removed, green added) before anything changes.

Safe, YOLO, and plan modes ​

Safe follows your configured approval settings.

YOLO auto-approves most mutating tools until you switch back. Recursive folder deletes still ask, and a per-tool deny still wins.

/plan enters sticky read-only planning mode. It blocks mutations until you leave. To leave, click the Plan badge in chat or run /config to switch back to Safe or YOLO.

Run /compact [instructions] to summarize older turns on demand. The optional instructions are passed to the compaction request, so important details survive the summary.

Memory ​

Add long-term memories with /memory add [kind] [scope] <text>. Kinds are preference, fact, instruction, or summary. Scopes are global or vault.

Review stored memories with /memory review. Export them with /memory export. Clear everything with /memory clear --confirm.

The agent retrieves memories explicitly through the search_memory tool. The plugin never injects them automatically.

Semantic indexing ​

Run /semantic-index start to build a vector index over your current scope (vault, folder, tag, or the active note's folder). A full-vault build requires /semantic-index start vault --confirm-vault. 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>, notes with /todo note <id> <note>, and retitle the plan with /todo title <name>.

The plan tracker panel shows live progress.

Real-time controls ​

While the agent streams a response, you can:

  • /steer [text] steers the current turn mid-stream.
  • /follow-up [text] queues a follow-up message behind the active turn.
  • /redirect [text] stops the active turn and answers the new text instead.

Export conversations ​

Run /export to save the current conversation as a Markdown note in the vault.

Reasoning effort ​

Set your reasoning effort with /effort [level] (for example, /effort high). The choice persists across messages. Change it with /effort again or with the effort knob in the composer.

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.

Released under the MIT License.