How juandiegohernandezangarita13-png/mmae-agent works
MMAE/JEV is a local and distributed autonomous-agent platform. Its documented core workflow accepts a goal through an interface, protects exact literals, creates a macro plan, expands it into a detailed action, then executes and reflects on tool results. JEV adds a broader agent runtime, skills, memory, verification, multimodal tools, and desktop/remote interfaces. Edges follow README statements or references in excerpted/indexed files; unsampled subsystem wiring is omitted or marked unverified.
Main entry points: agent.py JEV agent · base.py Tool registry · micro_engine.py Micro engine · planner.py Task planner
Components 23
User interfaces
- CLI
main.py - Desktop app
app.py - Telegram bot
jev/interfaces/telegram_bot.py
Agent reasoning
- JEV agent
jev/core/agent.py - Literal bypass
utils/bypass.py - Macro engine
core/macro_engine.py - Task planner
core/planner.py - Micro engine
core/micro_engine.py - Reflection loop
core/reflection_loop.py - LLM client
jev/core/llm_client.py
Execution and skills
- Tool registry
jev/tools/base.py - Skill manager
jev/skills/manager.py - Muse executor
jev/core/muse/execution_engine.py - Command sandbox
jev/core/sandbox.py
Memory and improvement
- Agent memory
jev/core/memory.py - Self improvement
jev/core/self_evolving_agent.py - Fine-tuning
jev/learning/finetune_engine.py
Specialized capabilities
- Claude Code bridge
jev/core/claude_code_bridge.py - Vision engine
jev/core/vision_engine.py - Image generator
jev/core/image_generator.py - Web sidecar
jev/ui/web_sidecar_server.py - Claude Code service
jev/core/claude_code_service.py
Other components
- User
Connections 28
- User → CLI: submits goal
- User → Desktop app: submits goal
- User → Telegram bot: submits goal
- CLI → JEV agent: invokes
- Desktop app → JEV agent: invokes
app.py - Telegram bot → JEV agent: invokes
- JEV agent → Task planner: plans tasks
jev/core/agent.py - JEV agent → LLM client: queries
jev/core/agent.py - JEV agent → Tool registry: dispatches tools
jev/core/agent.py - JEV agent → Agent memory: uses memory
jev/core/agent.py - JEV agent → Reflection loop: reflects
- JEV agent → Self improvement: delegates improvement
jev/core/agent.py - JEV agent → Skill manager: uses skills
jev/core/agent.py - JEV agent → Vision engine: uses vision
- JEV agent → Image generator: uses image tools
- JEV agent → Claude Code bridge: delegates coding
- JEV agent → Web sidecar: connects UI
app.py - Literal bypass → Macro engine: shields literals
core/macro_engine.py - Macro engine → Task planner: structures task
- Task planner → Micro engine: provides subtasks
- Literal bypass → Micro engine: restores literals
core/micro_engine.py - Micro engine → LLM client: fallback queries
core/micro_engine.py - Micro engine → Tool registry: emits tool actions
- Tool registry → Muse executor: executes action
- Muse executor → Command sandbox: validates commands
jev/core/muse/execution_engine.py - Skill manager → Tool registry: adapts skills
jev/skills/manager.py - Claude Code bridge → Claude Code service: forwards requests
jev/core/claude_code_bridge.py - Fine-tuning → Tool registry: exposes tool
jev/learning/finetune_engine.py