How simonbouhier/lyra_reborn works
Lyra Reborn is a local-LLM cognitive layer combining conversation, control, memory, and knowledge exploration. Its user workflow enters through the local app, runs a session turn through the cognitive loop and language model, and returns a response while updating session state and memory. The app also exposes a separate durable request/journal workflow. ESMM provides a distinct graph-exploration capability, and La Vigie is a separate campaign instrument. Some wiring is omitted because the sampled excerpts do not show those implementations.
Main entry points: session.py Session manager · loop.py Cognitive loop · main.py HTTP interface · orchestrator.py ESMM orchestrator
Components 22
Local app
- HTTP interface
app/main.py - Session manager
app/session.py - Dialogue mode
app/dialogue.py - Journal requests
app/requests.py - Journal client
app/static/journal-client.js - Session persistence
app/storage.py
Cognitive loop
- Cognitive loop
core/loop.py - Dynamic control
core/control/controller.py - Generation knobs
core/knobs.py - LLM clients
core/llm.py
Memory and knowledge
- Knowledge graph
memory/graph/store.py - Memory ecology
memory/ecology/ecology.py - Case memory
memory/cbr/memento.py - Embeddings
core/embeddings.py
Graph exploration
- ESMM orchestrator
explore/esmm/orchestrator.py - Gap detection
explore/esmm/gaps.py - Proposal consensus
explore/esmm/consensus.py
La Vigie
- Vigie campaign
agency/tools/vigie/campaign.py - Vigie campaign V2
agency/tools/vigie/campaign_v2.py
Other components
- User
- Local Ollama
- Ollama embeddings
Connections 18
- User → HTTP interface: sends requests
- HTTP interface → Session manager: handles turns
- Session manager → Dialogue mode: restores dialogue
- Session manager → Cognitive loop: runs turn
- Cognitive loop → Dynamic control: updates control
- Cognitive loop → Generation knobs: uses settings
- Cognitive loop → LLM clients: generates text
- LLM clients → Local Ollama: calls API
- Session manager → Knowledge graph: updates concepts
- Session manager → Memory ecology: triages turn
- Session manager → Session persistence: saves state
- Session manager → Session persistence: loads state
- Embeddings → Ollama embeddings: requests vectors
- HTTP interface → Journal requests: dispatches journal
- Journal client → HTTP interface: calls API
- ESMM orchestrator → Gap detection: detects gaps
- ESMM orchestrator → Knowledge graph: reads and commits
- ESMM orchestrator → Proposal consensus: builds consensus