Stop watching your agents rediscover the same solutions every session. Sibyl gives them persistent memory, semantic search, and the ability to learn from each other.
Every time you start a new Claude Code session, your agent forgets everything. That OAuth gotcha you debugged for 2 hours? Gone. The pattern that finally made your tests pass? Vanished. The configuration quirk that took forever to figure out? Lost to the void.
Your AI agents are brilliant but amnesiac.
Sibyl is a Collective Intelligence Runtime—a persistent knowledge graph that gives your AI agents:
┌──────────────────────────────────────────────────────────────────┐
│ Your AI Agents │
│ Claude Code • Custom Agents • Multi-Agent Systems │
└────────────────────────────┬─────────────────────────────────────┘
│
┌────────────────────┼────────────────────┐
│ │ │
▼ ▼ ▼
┌─────────┐ ┌───────────┐ ┌──────────┐
│ Skills │ │ Hooks │ │ CLI │
│ Teach │ │ Inject │ │ Express │
│ workflow│ │ context │ │ power │
└─────────┘ └───────────┘ └──────────┘
│ │ │
└────────────────────┼────────────────────┘
│
▼
┌──────────────────────────────────────────────────────────────────┐
│ Sibyl Server │
│ Knowledge Graph (FalkorDB) • Semantic Search • Task Management │
└──────────────────────────────────────────────────────────────────┘
│
▼
┌──────────────────────────────────────────────────────────────────┐
│ Web UI │
│ Human collaboration • Project management • Graph explorer │
└──────────────────────────────────────────────────────────────────┘Skills teach your agents structured workflows:
# The agent learns to search before implementing
sibyl search "authentication patterns"
# Track work with full lifecycle
sibyl task start task_xyz
# Capture learnings when done
sibyl task complete task_xyz --learnings "OAuth tokens need refresh..."Hooks automatically inject relevant knowledge into every prompt—your agent doesn't even need to ask.
Web UI for collaboration and oversight:
CLI for power users and scripting:
# Install the CLI
pip install sibyl-cli
# Start the infrastructure
docker run -d --name falkordb -p 6380:6379 falkordb/falkordb:latest
# Configure and start
cd sibyl && moon run dev
# Search your knowledge
sibyl search "authentication patterns"
# Add a learning
sibyl add "Redis insight" "Connection pool must be >= concurrent requests"
# Manage tasks
sibyl task list --status doing
sibyl task complete task_xyz --learnings "OAuth tokens expire after 1 hour"Sibyl enforces a simple but powerful cycle:
┌─────────────────────────────────────────────────────┐
│ 1. RESEARCH │
│ Search for existing patterns before coding │
│ sibyl search "what you're implementing" │
└─────────────────────────┬───────────────────────────┘
▼
┌─────────────────────────────────────────────────────┐
│ 2. DO │
│ Work on your task with context from search │
│ sibyl task start task_xyz │
└─────────────────────────┬───────────────────────────┘
▼
┌─────────────────────────────────────────────────────┐
│ 3. REFLECT │
│ Capture learnings for future agents │
│ sibyl task complete task_xyz --learnings "..." │
└─────────────────────────────────────────────────────┘Every completed task makes your knowledge graph smarter. Every pattern discovered helps future agents work faster. The system learns as you work.
| Without Sibyl | With Sibyl |
|---|---|
| Agent rediscovers same solutions | Agent finds existing patterns instantly |
| Context lost between sessions | Knowledge persists forever |
| Manual prompting required | Hooks inject context automatically |
| No task tracking | Full lifecycle with learnings capture |
| Scattered documentation | Searchable, connected knowledge graph |
Built for AI agents that deserve to remember.