+--------------------------------------+
| LegionIO v1.4.108 |
| CLI / REST API / MCP / Chat |
+-----------------+--------------------+
|
+----------+----------+-------+--------+----------+----------+
| | | | | | |
transport crypt data cache settings llm gaia
(RabbitMQ) (Vault) (Sequel) (Redis) (config) (ruby_llm) (tick)
| | | | | | |
+----------+----------+-------+--------+----------+----------+
|
+-------------------------+-------------------------+
| | |
18 Core LEXs 13 Cognitive Domains 29 Service LEXs
(tasker, synapse, (234 sub-modules: (slack, redis, http,
scheduler, node, memory, emotion, ssh, s3, vault,
conditioner...) trust, prediction...) github, chef...)
| Library | Version | Purpose |
|---|---|---|
| legion-transport | 1.2.6 | RabbitMQ AMQP messaging with clustering, TLS, spool buffer |
| legion-crypt | 1.4.4 | Encryption, HashiCorp Vault, JWT, multi-cluster Vault |
| legion-data | 1.4.8 | Database persistence (SQLite/PostgreSQL/MySQL via Sequel) |
| legion-cache | 1.3.3 | Two-tier caching (Redis/Memcached with local fallback) |
| legion-settings | 1.3.8 | Config management, schema validation, secret resolver, DNS bootstrap |
| legion-llm | 0.3.11 | LLM integration with three-tier model escalation |
| legion-gaia | 0.9.8 | Cognitive coordination (tick cycle, dream cycle, channels) |
| legion-mcp | 0.4.0 | MCP server with Tier 0 behavioral intelligence routing |
| legion-rbac | 0.2.3 | Role-based access control (Vault-style flat policies) |
| legion-tty | 0.4.29 | Terminal UI (prompts, tables, spinners, AI chat shell) |
| legion-logging | 1.2.5 | Console + structured JSON logging + SIEM export |
| legion-json | 1.2.0 | JSON serialization (multi_json) |
Every agent runs a 13-phase cognitive loop every tick:
| Phase | Name | What Happens |
|---|---|---|
| 1 | Sensory Processing | Perceive environment, ingest new inputs |
| 2 | Emotional Evaluation | Evaluate inputs against emotional state |
| 3 | Memory Retrieval | Search episodic and semantic memory for relevant context |
| 4 | Knowledge Retrieval | Query Apollo shared knowledge store |
| 5 | Identity Entropy Check | Verify agent identity coherence |
| 6 | Working Memory Integration | Combine inputs, memories, knowledge into working state |
| 7 | Procedural Check | Check for matching procedural patterns (habits) |
| 8 | Prediction Engine | Generate predictions, compare with reality |
| 9 | Mesh Interface | Coordinate with other agents in the mesh |
| 10 | Gut Instinct | Fast heuristic evaluation (somatic markers) |
| 11 | Action Selection | Choose action based on all preceding phases |
| 12 | Memory Consolidation | Store new experiences in memory |
| 13 | Post-Tick Reflection | Metacognitive review of the tick |
When an agent goes idle, it enters a 7-phase dream cycle:
| Phase | Name | What Happens |
|---|---|---|
| 1 | Memory Audit | Review recent memories for importance |
| 2 | Association Walk | Find connections between unrelated memories |
| 3 | Contradiction Resolution | Resolve conflicting beliefs or memories |
| 4 | Agenda Formation | Form new priorities based on consolidated knowledge |
| 5 | Consolidation Commit | Persist reorganized memories |
| 6 | Dream Reflection | Evaluate what the dream cycle produced |
| 7 | Dream Narration | Generate a natural-language dream journal |
234 cognitive modules across 13 domain gems:
| Domain | Modules | What It Models |
|---|---|---|
| Executive | 23 | Planning, control, inhibition, working memory, decision-making |
| Attention | 24 | Spotlight, switching, salience, gating |
| Memory | 18 | Encoding, storage, retrieval, consolidation, decay |
| Affect | 17 | Emotion, mood, empathy, somatic markers, reward |
| Inference | 27 | Prediction, causation, belief updating |
| Social | 17 | Theory of mind, cooperation, trust, moral reasoning |
| Self | 16 | Metacognition, identity, self-model, narrative, personality |
| Learning | 14 | Habit, reinforcement, procedural learning, adaptation |
| Language | 9 | Inner speech, narrative, frame semantics |
| Imagination | 17 | Creativity, dreaming, mental simulation, prospection |
| Homeostasis | 20 | Balance, rhythm, energy, fatigue recovery |
| Defense | 15 | Bias detection, error monitoring, cognitive immune system |
| Integration | 17 | Cross-modal binding, coherence, Global Workspace Theory |
Extensions are Ruby gems named lex-*, auto-discovered at boot via Bundler.load.specs.
Each extension defines:
# Scaffold a new extension
legion lex create my_extension
# Add components
legion generate runner my_runner
legion generate actor my_actor
legion generate tool my_tool
Deterministic startup order:
Logging -> Settings -> Crypt -> Transport -> Cache -> Data
-> RBAC -> LLM -> GAIA -> Telemetry -> Extensions -> API
GAIA boots before extensions so the tick cycle and channel abstraction are ready when cognitive extensions wire their phase handlers.
Two-phase extension loading: all extensions require + autobuild first, then hook_all_actors starts AMQP timers and subscriptions.
Shutdown is the reverse order.
# CLI — 60+ commands, every one supports --json
legion start
legion task run http.request.get url:https://example.com
legion dashboard
# Interactive AI Chat
legion chat
# REST API (Sinatra + Puma)
curl http://localhost:4567/api/v1/tasks
# MCP Server (Model Context Protocol)
legion mcp
| Layer | Component | Role |
|---|---|---|
| Bones | lex-tasker | Raw task execution and chaining |
| Nerves | lex-synapse | Confidence-scored routing, autonomy levels, auto-revert |
| Mind | GAIA + Apollo | Dream replay, knowledge promotion, shared memory |
Autonomy scales with confidence: Observe (0-0.3) -> Filter (0.3-0.6) -> Transform (0.6-0.8) -> Autonomous (0.8-1.0).
Durable knowledge store for the cognitive mesh. PostgreSQL + pgvector.