From first launch to the agent loop.
Hetu docs are organised as “use it → understand it”: get notes and Code sessions working first, then read how retrieval, compression, permissions and workflows are built. Every page covers the UI path, configuration, where data lands and how to troubleshoot.
The five-minute path
Install and launch
Use a desktop installer or run from source; confirm the frontend and backend both respond.
Getting started →Connect a model
Add a provider and models, then set per-scenario defaults for chat, coding, Wiki and more.
Models & usage →Write your first note
Polish or expand with AI, generate an index, and give the knowledge base something to retrieve.
Notes →Run a Code session
Pick a repository, open a worktree, and let the agent read code, edit files and commit changes.
Code sessions →Feature map
Find the entry point for what you are doing. Each feature page includes the UI path, key settings and common troubleshooting.
Write and ask
- NotesDual editors, AI actions, history, sharing, export
- Agent workspaceSessions and topics, permission modes, queue, steering
- Theme & appearancePalettes, canvas backgrounds, acrylic material
- Models & usageProviders, scenario defaults, proxy, cost
Do the work
- Code sessionsWorktrees, branches, seven panels, PRs
- Kanban & automationTask board, workflows, background and scheduled jobs
- WhiteboardBoards, drawing tools, export and presentation
- Extensions & MCPAgents, skills, tools, MCP servers
Knowledge
- Knowledge baseChunking, vector index, coverage, semantic search
- Graph & WikiEntity extraction, generated project docs, automatic knowledge base and graph indexing
- Memories & DreamScopes, recall, consolidation and forgetting
Platform and data
Deep dives
For people who read or change the code: key mechanisms, defaults, guards, fallbacks and how to debug them.
- Agent loopSix steps, iteration limits, steering injection
- Permission modes & approvalsFive modes, three-layer decision, rule matching
- Context & auto compactionUsage algorithm, trigger ratio, summary retry
- Compression pipelineNode order, summary thresholds, config merge
- SSE event protocolFrame shapes, event list, stream differences
- Chunking & vector indexChunk parameters, virtual table writes, dimensions
- Semantic retrievalTwo recall paths, merge, consumers
- Knowledge graph extractionJSON contract, normalization, failure paths
- Wiki generationCollection, planning, concurrent writing
- Memory write & recallThree write paths, scoring formula
- Dream consolidationClustering, merging, decay and forgetting
- Task board & automationState machine, triggers, comment flow
- Workflow engineGraph validation, ten node types, human review
- Background & scheduled tasksDedupe, recovery, Cron
- Inbox & notificationsAggregation, merge actions, state semantics
Stuck?
http://localhost:5000/api/health first, then the logs/ directory under your data folder.
These pages are maintained alongside the code. If a page disagrees with the implementation, open a GitHub issue with the page and section.
