How upsonic/gpt-computer-assistant works
Upsonic is a Python framework for building autonomous and traditional AI agents. The primary workflow is a developer-created task submitted to an agent, which uses a selected language model and can invoke tools to produce a result. The repository also documents a separate layered OCR workflow and includes substantial subsystems for agent orchestration, model/provider support, tools, knowledge and memory, and integrations. Arrows are limited to relationships stated in the README or visible in the SOURCE INDEX; many implementation details were not sampled.
Main entry points: agent.py Traditional agent · tasks.py Task · __init__.py Model interface · ocr.py OCR interface
Components 22
Agent runtime
- Task
src/upsonic/tasks/tasks.py - Traditional agent
src/upsonic/agent/agent.py - Autonomous agent
src/upsonic/agent/autonomous_agent/autonomous_agent.py - Execution pipeline
src/upsonic/agent/pipeline/manager.py - Run results
src/upsonic/run/agent/output.py - Agent teams
src/upsonic/team/team.py
Models and tools
- Model interface
src/upsonic/models/__init__.py - OpenAI model
src/upsonic/models/openai.py - Model settings
src/upsonic/models/settings.py - Tool execution
src/upsonic/tools/builtin_tools.py - MCP tools
src/upsonic/tools/mcp.py
Knowledge and state
- Knowledge base
src/upsonic/knowledge_base/knowledge_base.py - Document loaders
src/upsonic/loaders/json.py - Agent memory
src/upsonic/memory/memory.py - Storage interface
src/upsonic/storage/base.py
Extensions
- OCR interface
src/upsonic/ocr/ocr.py - Document preparation
src/upsonic/ocr/layer_0/document_converter.py - OCR engines
src/upsonic/ocr/layer_1/engines - Messaging interfaces
src/upsonic/interfaces/manager.py - Graph workflows
src/upsonic/graph/graph.py - Usage tracking
src/upsonic/usage.py
Other components
- Developer
Connections 24
- Developer → Traditional agent: creates agent
README.md - Developer → Task: defines task
README.md - Developer → Autonomous agent: creates agent
README.md - Traditional agent → Task: runs task
README.md - Autonomous agent → Task: runs task
README.md - Traditional agent → Model interface: uses model
src/upsonic/agent/agent.py - Traditional agent → Execution pipeline: executes through
src/upsonic/agent/agent.py - Execution pipeline → Run results: produces results
src/upsonic/agent/pipeline/manager.py - Traditional agent → Tool execution: executes tools
src/upsonic/agent/agent.py - Task → Tool execution: defines tools
src/upsonic/tasks/tasks.py - Model interface → OpenAI model: supports provider
src/upsonic/models/openai.py - Model interface → Model settings: uses settings
src/upsonic/models/__init__.py - Tool execution → MCP tools: extends with MCP
- Task → Knowledge base: can use knowledge
src/upsonic/tasks/tasks.py - Knowledge base → Document loaders: loads documents
src/upsonic/knowledge_base/knowledge_base.py - Traditional agent → Agent memory: uses memory
src/upsonic/agent/agent.py - Agent memory → Storage interface: persists through
- Developer → OCR interface: requests text extraction
README.md - OCR interface → Document preparation: prepares documents
README.md - OCR interface → OCR engines: runs engine
README.md - Developer → Messaging interfaces: connects interfaces
- Developer → Graph workflows: builds workflows
- Traditional agent → Agent teams: coordinates agents
- Traditional agent → Usage tracking: records usage
src/upsonic/agent/agent.py