How ayr1k/luma works
Luma is a Windows desktop AI agent for conversations and project work. Its main workflow is user input through the local client API, asynchronous task execution by AgentCore, model inference and permission-bounded tool use, then persisted results for review. Separate capabilities cover context management, extensions, project/Git workflows, and file previews. The README documents additional UI and product behavior beyond the sampled excerpts; the graph uses documented relationships where implementation wiring was not shown.
Main entry points: service.py Local API · core.py Agent Core · tools.py Tool execution · builtin_tools.py Built-in tools
Components 23
Client and API
- Desktop client
local_agent/desktop.py - Local API
local_agent/service.py - Model settings
local_agent/preferences.py - Model profiles
local_agent/model_profiles.py
Agent execution
- Task runner
local_agent/tasks.py - Agent Core
local_agent/core.py - Inference client
local_agent/model.py - Tool execution
local_agent/tools.py - Built-in tools
local_agent/builtin_tools.py - Attachments
local_agent/attachments.py
Project operations
- Session storage
local_agent/storage.py - Project files
local_agent/files.py - Web search
local_agent/search.py - Git operations
local_agent/git_manager.py - Change preview
local_agent/code_preview.py
Extensions and context
- Context management
local_agent/context.py - Extensions and Skills
local_agent/extensions.py - Plugin worker
local_agent/plugin_worker.py - Preset plugins
preset_plugins/luma-project-overview/main.py
Other components
- User
- Inference service
- Local workspace
- Git remote
Connections 24
- User → Desktop client: uses
- Desktop client → Local API: requests
- Local API → Model settings: configures
- Local API → Model profiles: manages
- Local API → Task runner: starts tasks
- Task runner → Agent Core: runs action
- Agent Core → Inference client: requests inference
- Inference client → Inference service: sends prompts
- Agent Core → Tool execution: dispatches tools
- Tool execution → Built-in tools: dispatches built-ins
- Agent Core → Attachments: prepares input
- Agent Core → Session storage: persists session
- Local API → Session storage: reads projects
- Agent Core → Context management: manages context
- Built-in tools → Project files: uses file tools
- Built-in tools → Web search: searches web
- Tool execution → Extensions and Skills: resolves extensions
- Extensions and Skills → Plugin worker: runs code plugin
- Plugin worker → Preset plugins: executes entry
- Project files → Local workspace: reads and writes
- Git operations → Local workspace: inspects workspace
- Local API → Git operations: serves Git actions
- Git operations → Git remote: fetches and pushes
- Local API → Change preview: provides previews