Cross-Model Continuity
Switch between Claude Code, Codex, Gemini CLI, and Cursor without losing context.
The problem
Every AI coding assistant starts from zero. Switch from Claude to Codex and you lose your task list, architecture decisions, and session state. Delimit fixes this with two persistence layers that work across all models.
How it works
Ledger
Task tracker at ~/.delimit/ledger/. Every model reads and writes to the same ledger. Tasks persist across sessions, reboots, and model switches.
Memory
Project knowledge at MEMORY.md and topic files. Architecture decisions, conventions, and context that any model can reference.
The handoff
Here is what a real handoff looks like across three models:
Builds auth module. Adds LED-047 to ledger: "auth module complete, needs rate limiting." Writes session handoff with blockers.
Starts session. Reads ledger, sees LED-047. Picks up rate limiting. Updates ledger to "in progress." Runs tests. Marks done.
Reviews the full ledger history. Sees auth + rate limiting are done. Runs governance checks. Flags a missing test. Adds new ledger item.
What syncs across models
| Data | Location | Purpose |
|---|---|---|
| Ledger items | ~/.delimit/ledger/ | Tasks, priorities, status |
| Memory index | MEMORY.md | Architecture decisions, conventions |
| Session handoffs | ~/.delimit/sessions/ | What was done, what's next, blockers |
| Governance state | .delimit/ | Policies, audit history, evidence |
Supported models
Any tool that supports the Model Context Protocol can connect to Delimit's MCP server and share the same state.