CitadelBring Your Own AI
Your AI forgets everything. Here's the fix.
Keep your own LLM. Keep your own infrastructure. Citadel just gives it memory that persists across sessions — through an MCP connection called drawbridge.
Your AI
Claude Code, GPT, or any LLM that supports MCP tool connections
Your infrastructure
Your Linux machine, your network, your data stays yours
What Citadel adds
Persistent memory — plans, decisions, context, relationships that survive every session
What drawbridge is
An MCP server. Your AI calls tools to read and write to Citadel. That's the whole integration.
Connected to Citadel via drawbridge. Loading: 3 active outcomes, 12 tasks, 8 decisions from prior sessions. Last session: 14 hours ago.
Pick up where we left off.
Last session you finalized the API schema and assigned the frontend to your Wednesday block. The migration script is done — tests passed. The only open item is the staging deploy, which is ready to trigger.
Wait. I haven't told you any of that. This is a fresh session.
It is a fresh session. But I loaded your working state through drawbridge when I started — the decisions, task status, and test results from last time. You don't have to re-brief me. That's the point.
So every session from now on starts like this?
Every session. And anything I learn today gets written back. Tomorrow's session will know what we did tonight.
What Survives
continuity after the session ends
Normal AI session
starts cold, context lost on close
reset
Notes + docs
scattered, unlinked, not queryable
fragmented
Citadel + drawbridge
structured memory, loaded automatically
preserved
The problem
Right now you're managing context by hand. You paste old conversations, re-explain your codebase, and hope the AI infers what matters. When the session ends, everything it learned disappears.
The difference
With drawbridge, your AI reads structured context at session start — decisions, plans, relationships, open work. It writes back findings as it works. Next session starts where the last one left off. Work compounds instead of resetting.