How tencent/loopforge works
LoopForge is a self-hosted coding-agent workspace with two user paths: interactive Main Bot conversations and repeatable SOP runs. Both connect work to requirements, execute agents in isolated workspaces, and deliver reviewable code or artifacts. The README documents these product relationships; the source excerpts are incomplete, so internal call paths are shown only where documentation supports them, not inferred from directory names.
Main entry points: engine.go SOP engine · httpapi HTTP API · useChat.js Chat session · workflow_graph.go SOP node stages
Components 19
User experience
- Chat workbench
chatbot/src/App.jsx - Chat session
chatbot/src/useChat.js - Run and artifact views
chatbot/src/ArtifactPanel.jsx
Delivery workflows
- Requirements
internal/domain/requirement.go - SOP engine
internal/engine/engine.go - SOP node stages
internal/engine/workflow_graph.go - Human approval gates
internal/engine/gate_submit.go
Agent execution
- Coding-agent runtime
internal/runtime/runtime.go - Sandbox manager
internal/runtime/sandbox_manager.go - Remote runner
internal/runner/daemon.go - Conversation workspace
internal/engine/conversation_workspace.go
Platform services
- HTTP API
internal/httpapi - Platform database
internal/repo/repo.go - Run observability
internal/engine/engine_observe.go - Notifications
internal/notification/service.go
Results and integrations
- Git hosting
- Artifacts and logs
internal/engine/artifacts.go - Issue and PR tracker
internal/tracker/service.go
Other components
- Team member
Connections 24
- Team member → Chat workbench: starts chat
README.md - Chat workbench → HTTP API: sends requests
README.md - HTTP API → Chat session: opens conversation
- Chat session → Conversation workspace: requests workspace
README.md - Conversation workspace → Sandbox manager: opens sandbox
README.md - Conversation workspace → Coding-agent runtime: calls agent
README.md - Coding-agent runtime → Remote runner: dispatches work
- Team member → HTTP API: starts SOP
README.md - HTTP API → SOP engine: starts run
- SOP engine → SOP node stages: runs stages
README.md - SOP node stages → Coding-agent runtime: runs agent nodes
README.md - SOP node stages → Human approval gates: pauses for approval
README.md - SOP engine → Sandbox manager: isolates execution
README.md - Chat session → Requirements: links requirement
README.md - SOP engine → Requirements: tracks delivery
README.md - HTTP API → Platform database: persists platform data
- SOP engine → Artifacts and logs: records outputs
README.md - SOP engine → Run observability: tracks run state
README.md - SOP engine → Git hosting: opens PR or MR
README.md - Chat session → Git hosting: delivers code
README.md - HTTP API → Issue and PR tracker: integrates issues
- SOP engine → Notifications: sends updates
- HTTP API → Run and artifact views: serves run results
- Run and artifact views → Team member: shows artifacts