How tridz-dev/huf works
HUF is a self-hosted AI infrastructure platform with a web interface and a Frappe-based backend. A user configures an agent or workflow, then invokes it through chat or the public API; the runtime manages conversation state, calls an LLM provider, and can use tools or grounded knowledge. Separate capability areas cover visual flows, integrations and messaging, reusable apps, and governance. The graph combines documented product capabilities with source-index references and excerpts; links to unsampled subsystems are included only where the README documents the relationship.
Main entry points: agent_integration.py Agent runtime · router.py Public API router · tool_registry.py Tool registry · context.py Request context
Components 22
User access
- HUF web app
frontend/src/App.tsx - Public API router
huf/api/v1/router.py - Request context
huf/api/v1/context.py
Agent runtime
- Agent runtime
huf/ai/agent_integration.py - Conversation state
huf/ai/conversation_manager.py - LLM provider
huf/ai/providers/litellm.py - Context assembly
huf/ai/context_segments.py - Cost analytics
huf/ai/cost_calculator.py - Run budgets
huf/ai/run_budget.py
Workflow automation
- Visual flow engine
huf/ai/flow_engine.py - Agent orchestration
huf/ai/orchestration/orchestrator.py
Tools and grounding
- Tool registry
huf/ai/tool_registry.py - Built-in tools
huf/ai/tools/_registry.py - MCP client
huf/ai/mcp_client.py - Knowledge retrieval
huf/ai/knowledge/retriever.py - Document extraction
huf/ai/ocr_engine.py - Apps ecosystem
huf/ai/app_seeding/apps_loader.py - Messaging gateways
huf/ai/tools/telegram_webhook.py
Platform governance
- Capability permissions
huf/permissions.py - Agent access checks
huf/ai/agent_access.py - Transaction control
huf/ai/transaction.py
Other components
- User
Connections 23
- User → HUF web app: uses
README.md - User → Public API router: invokes
README.md - Public API router → Request context: builds context
huf/api/v1/router.py - Public API router → Agent runtime: dispatches run
- Agent runtime → Conversation state: manages history
huf/ai/agent_integration.py - Agent runtime → LLM provider: requests completion
huf/ai/agent_integration.py - Agent runtime → Context assembly: builds context
huf/ai/agent_integration.py - Agent runtime → Run budgets: applies budgets
huf/ai/agent_integration.py - LLM provider → Cost analytics: tracks cost
huf/ai/providers/litellm.py - Agent runtime → Tool registry: invokes tools
huf/ai/agent_integration.py - Tool registry → Capability permissions: checks capabilities
huf/ai/tool_registry.py - Agent runtime → MCP client: uses MCP
huf/ai/agent_integration.py - Agent runtime → Knowledge retrieval: grounds responses
README.md - Agent runtime → Document extraction: extracts documents
huf/ai/agent_integration.py - Visual flow engine → Agent runtime: runs agent nodes
huf/ai/flow_engine.py - Agent orchestration → Agent runtime: coordinates agents
README.md - Messaging gateways → Agent runtime: starts agent runs
huf/ai/tools/telegram_webhook.py - Apps ecosystem → Tool registry: packages capabilities
README.md - Request context → Agent access checks: authorizes agent
huf/api/v1/endpoints/responses_stream.py - Agent access checks → Capability permissions: checks roles
- Agent runtime → Transaction control: commits background state
huf/ai/agent_integration.py - User → Visual flow engine: designs workflows
README.md - User → Apps ecosystem: reuses skills
README.md