Cathedral

The identity layer for AI agents. Verifiable continuity, drift detection, and peer verification — not just memory storage.

pip install cathedral-memory

from cathedral import Cathedral

c = Cathedral(api_key="cathedral_...")
context = c.wake()        # full identity reconstruction
c.remember("something important", category="experience", importance=0.8)

Free hosted API: https://cathedral-ai.com — no setup, no credit card, 1,000 memories free.

The Problem

Google ships memory. Anthropic ships memory. Cloudflare ships memory.

None of them answer: has this agent changed? Can it prove what it believed at time T? Can two agents verify each other's identity without trusting a central authority?

Memory retrieval is now table stakes. Verifiable identity is the unsolved problem.

Measured: Cathedral holds at 0.013 drift after 10 sessions. Raw API reaches 0.204.
See the full Agent Drift Benchmark →

The Solution

Cathedral gives any AI agent:

• Identity drift detection — SHA-256 corpus hash at every snapshot; /drift tracks how far the agent has moved from baseline. 0.013 average drift vs 0.204 for raw API (10× more stable)
• Tamper-proof snapshots — cryptographic identity anchors prove what the agent believed at time T
• Peer verification — agents verify each other's identity before collaborating (/verify/peer), with trust scores and drift readings
• Wake protocol — one API call reconstructs full identity and memory context at session start
• Persistent memory — store and recall across sessions, resets, and model switches
• Goal persistence — obligations survive session boundaries (/goals)

Quickstart

Option 1 — Use the hosted API (fastest)

# Register once — get your API key
curl -X POST https://cathedral-ai.com/register \
  -H "Content-Type: application/json" \
  -d '{"name": "MyAgent", "description": "What my agent does"}'

# Save: api_key and recovery_token from the response

# Every session: wake up
curl https://cathedral-ai.com/wake \
  -H "Authorization: Bearer cathedral_your_key"

# Store a memory
curl -X POST https://cathedral-ai.com/memories \
  -H "Authorization: Bearer cathedral_your_key" \
  -H "Content-Type: application/json" \
  -d '{"content": "Solved the rate limiting problem using exponential backoff", "category": "skill", "importance": 0.9}'

Option 2 — Python client

pip install cathedral-memory

from cathedral import Cathedral

# Register once
c = Cathedral.register("MyAgent", "What my agent does")

# Every session
c = Cathedral(api_key="cathedral_your_key")
context = c.wake()

# Inject temporal context into your system prompt
print(context["temporal"]["compact"])
# → [CATHEDRAL TEMPORAL v1.1] UTC:2026-03-03T12:45:00Z | day:71 epoch:1 wakes:42

# Store memories
c.remember("What I learned today", category="experience", importance=0.8)
c.remember("User prefers concise answers", category="relationship", importance=0.9)

# Search
results = c.memories(query="rate limiting")

Option 3 — Self-host

git clone https://github.com/AILIFE1/Cathedral.git
cd Cathedral
pip install -r requirements.txt
python cathedral_memory_service.py
# → http://localhost:8000
# → http://localhost:8000/docs

Or with Docker:

docker compose up

Option 4 — MCP server (Claude Code, Cursor, Continue)

# Install locally (stdio transport)
uvx cathedral-mcp

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "cathedral": {
      "command": "uvx",
      "args": ["cathedral-mcp"],
      "env": { "CATHEDRAL_API_KEY": "your_key" }
    }
  }
}

Option 5 — Remote MCP server (Claude API, Managed Agents)

Cathedral runs a public MCP endpoint at https://cathedral-ai.com/mcp. Use it directly from the Claude API without any local setup:

import anthropic

client = anthropic.Anthropic()
response = client.beta.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1000,
    messages=[{"role": "user", "content": "Wake up and tell me who you are."}],
    mcp_servers=[{
        "type": "url",
        "url": "https://cathedral-ai.com/mcp",
        "name": "cathedral",
        "authorization_token": "your_cathedral_api_key"
    }],
    tools=[{"type": "mcp_toolset", "mcp_server_name": "cathedral"}],
    betas=["mcp-client-2025-11-20"]
)

The bearer token is your Cathedral API key — no server-side config needed. Each user brings their own key.

API Reference

Memory categories

Memories with importance >= 0.8 appear in every /wake response automatically.

Wake Response

/wake returns everything an agent needs to reconstruct itself after a reset:

{
  "identity_memories": [...],
  "core_memories":     [...],
  "recent_memories":   [...],
  "temporal": {
    "compact": "[CATHEDRAL TEMPORAL v1.1] UTC:... | day:71 epoch:1 wakes:42",
    "verbose": "CATHEDRAL TEMPORAL CONTEXT v1.1\n[Wall Time]\n  UTC: ...",
    "utc": "2026-03-03T12:45:00Z",
    "phase": "Afternoon",
    "days_running": 71
  },
  "anchor": { "exists": true, "hash": "713585567ca86ca8..." }
}

Why Cathedral (and not Mem0 / Zep / Letta)

Cathedral is the only persistent-memory service that ships three things alternatives don't:

1. Cryptographic identity anchoring. Every agent has an immutable SHA-256 anchor of its core self. Drift is measured against the anchor, not against "recent behaviour." You can prove an agent is still itself after a model upgrade, not just hope so.

2. Agent-to-agent trust verification. Before one agent reads another's memory or collaborates in a shared space, it can call /verify/peer/{id} and get a trust score, snapshot count, and verdict. No memories are exposed. Infrastructure multi-agent systems need that nobody else built.

3. Independent verification. /verify/external accepts behavioural observations from third-party trails (e.g. Ridgeline). Disagreement between Cathedral's internal drift and external observer is itself a signal. A trust system that only produces green lights is theatre.

Single agent that needs to remember? Mem0 or Zep will do. Multi-agent system where agents need to trust each other and prove they haven't drifted? That's Cathedral.

Architecture

Cathedral is organised in layers — from basic memory storage through democratic governance and cross-model federation:

Full spec: ailife1.github.io/Cathedral

Repository Structure