How marksman8/hyper works
AURA is a local-first, model-agnostic personal assistant runtime. A user speaks or sends a task through a client; the authenticated API hands conversation turns to the orchestrator, which generates a response, persists conversation and memory, and can invoke permission-gated tools. Optional branches add voice, screen perception, desktop presence, evaluation, and human-gated configuration evolution. The README documents these capabilities; connections for unsampled subsystems are shown only where that documentation or the source index supports them.
Main entry points: orchestrator.py Turn orchestrator · app.py Local API · session.py Voice session · emitter.py Event bus
Components 23
Clients and API
- CLI clients
src/aura/clients/cli.py - Local API
src/aura/api/app.py
Conversation runtime
- Turn orchestrator
src/aura/core/orchestrator.py - Agent loop
src/aura/core/agent.py - Model providers
src/aura/models/providers/openai_compat.py - Tool execution
src/aura/tools/runner.py - Event bus
src/aura/core/emitter.py
Memory and security
- Persistent memory
src/aura/memory/store.py - Fact extraction
src/aura/memory/extraction.py - Access and approvals
src/aura/security/permissions.py - Audit trail
src/aura/security/audit.py
Voice and perception
- Voice session
src/aura/voice/session.py - Audio capture
src/aura/voice/audio/capture.py - Voice activity
src/aura/voice/vad.py - Core HTTP client
src/aura/voice/core_client.py - Screen perception
src/aura/vision/service.py - Desktop presence
src/aura/presence/app.py
Evaluation and evolution
- Evaluation harness
src/aura/eval/runner.py - Result comparison
src/aura/eval/compare.py - Config evolution
src/aura/evolution/search.py - n8n integration
src/aura/automation/n8n.py
Other components
- User
- Local model server
Connections 24
- User → CLI clients: requests tasks
README.md - CLI clients → Local API: calls API
README.md - User → Voice session: speaks
README.md - Voice session → Audio capture: captures audio
src/aura/voice/session.py - Voice session → Voice activity: detects speech
src/aura/voice/session.py - Voice session → Core HTTP client: requests replies
src/aura/voice/session.py - Core HTTP client → Local API: streams chat over HTTP
src/aura/voice/core_client.py - Local API → Turn orchestrator: routes turns
src/aura/api/app.py - Turn orchestrator → Model providers: generates responses
src/aura/core/orchestrator.py - Model providers → Local model server: uses model server
README.md - Turn orchestrator → Agent loop: runs bounded loop
src/aura/core/orchestrator.py - Agent loop → Tool execution: executes selected tools
README.md - Tool execution → Access and approvals: checks permissions
README.md - Turn orchestrator → Persistent memory: loads and persists turns
src/aura/core/orchestrator.py - Turn orchestrator → Fact extraction: learns facts
src/aura/core/orchestrator.py - Turn orchestrator → Audit trail: records actions
src/aura/core/orchestrator.py - Local API → Access and approvals: authenticates requests
src/aura/api/app.py - Local API → Persistent memory: serves session views
src/aura/api/app.py - Local API → Event bus: publishes events
src/aura/api/app.py - Screen perception → Event bus: emits privacy state
src/aura/vision/service.py - Event bus → Desktop presence: updates display
src/aura/presence/app.py - Evaluation harness → Result comparison: scores results
src/aura/clients/eval_cli.py - Config evolution → Result comparison: compares candidates
src/aura/evolution/search.py - n8n integration → User: supports integrations