Feature Map

Agentic Chat turns the Obsidian sidebar into an agent workspace rather than a plain chatbot.

Core chat

• Native Obsidian ItemView in the sidebar.
• Streaming assistant output.
• Inline tool-call step cards.
• Reasoning tokens in a collapsible section when the model provides them.
• Obsidian-style Markdown rendering, including callouts and Mermaid diagrams.
• Session tabs, history browsing, rename, delete, export, and clear actions.

Vault work

• Read and write notes.
• Edit notes by exact replacement.
• Search paths and note contents through the unified vault_inspect meta-tool.
• Traverse local note graph context.
• Read and update YAML frontmatter through Obsidian APIs.
• Rename notes with backlink preservation.
• Trash-safe file and empty-folder deletes.

Agent workflows

• Inline clarification prompts with ask_user.
• Safe and YOLO approval postures.
• Sticky read-only plan mode.
• Working directories for scoped auto-approval.
• Gitignore-style ignore list.
• /undo for the most recent vault mutation.
• Long-session compaction before context overflow, plus /compact [instructions] for manual compaction.
• Token, cost, next-request estimate, cost alert, and spend-cap readouts.
• Real-time turn steering with /steer, queued follow-ups with /follow-up, and hard redirects with /redirect.
• Reasoning effort control via /effort.

Memory

• Long-term memory with kinds: preference, fact, instruction, summary.
• Scopes: global, vault.
• Add memories with /memory add.
• Search memories with the search_memory tool.
• Export or clear memory stores.

Semantic retrieval

• Opt-in semantic index over vault notes.
• Configurable embedding provider (OpenRouter, Ollama, OpenAI-compatible).
• Scoped indexing by vault, folder, tag, or the active note's folder.
• /semantic-index commands to estimate, start, or cancel indexing.

Document import

• Import PDF, EPUB, DOCX, PPTX, and XLSX files into plugin-managed source artifacts.
• import_pdf and import_document tools extract text with provenance and citation anchors.
• Artifact chunks are inspectable with read_artifact and search_artifact.

Extensibility

• Agent plugins — packages in the vault (.agentic-plugins/) that follow the Agent Plugins 1.0.0 spec and contribute skills and MCP servers; single source of truth for both.
• Skills loaded from plugin SKILL.md files, alongside built-in skills; plugin skills load first and win name collisions.
• Built-in self-knowledge skill with tool inventory, doomloop guards, and error patterns.
• Subagent roles (single built-in Explorer, read-only recon; no custom roles).
• Built-in subagent: Explorer (single).
• Optional web search and URL fetch.
• Optional HTTPS Streamable HTTP MCP tools, generated into plugin packages from the settings UI.
• /doctor audit of agent plugin spec compliance.
• Optional OTLP or Langfuse observability export.
• Artifact system for large tool outputs.
• read_skill tool to load full skill content on demand.

Todo tracking

• /todo commands to track milestones, tests, and checkpoint commits.
• Integrates with the plan tracker panel for visual progress.