Agent-powered tools

Summary

Use the existing LLM integration and conversation infrastructure to build useful agentic workflows. Expose saved conversations through the graph, support branch/fork workflows, connect conversations to related entities (tasks, code, notes), and expose Space capabilities as MCP tools with risk-gated approval models.

Why

The LLM and agent infrastructure is already substantial (agent runner, preset system, multiple providers, MCP transport, conversation store). This goal is about making those pieces coherent and useful — connecting conversations to the graph, giving agents safe access to Space capabilities, and keeping LLM features subordinate to the entity model.

Success criteria

• Saved conversations are exposed through the graph and linked to related entities
• Branch/fork conversation workflows work cleanly
• Agent presets expose useful Space capabilities with clear risk boundaries
• MCP tool integration stays provider-agnostic

Features implementing this goal

• Agent Runner System — runtime agents with tool execution, presets, and approval flow

Bugs

(none yet)

Related

• Subsystems — Application Features (LLM & Conversations, Agent System, MCP Server)
• Milestones (Milestone 5)
• Project History — Phase VI: Agent layer (May 2026)
• Agent Layer Design
• Agent Presets