Skip to content
Gee-Code

Sessions & Continuity

Gee-Code preserves context across sessions through multiple mechanisms — session history, continuity ledgers, memory layers, and project activity logs. The AI always knows what happened before, even across restarts and long breaks.

Session Lifecycle

Section titled “Session Lifecycle”
  1. Start — launch gee-code or gee-code session-name in a project directory
  2. Work — interactive conversation with tool execution
  3. End — session transcript saved, memory facts extracted
  4. Resume — pick up where you left off with /resume

Named Sessions

Section titled “Named Sessions”

Run multiple independent conversations in the same project:

Terminal window
gee-code                    # Default session
gee-code auth-refactor      # Named session for auth work
gee-code bug-fix-422        # Named session for a specific bug

Each session has its own history, context files, and continuity ledger. They don’t interfere with each other.

Resuming Sessions

Section titled “Resuming Sessions”
/resume              # Show recent sessions, pick one to resume
/history             # View conversation history

When you resume, the full conversation transcript is restored — the AI sees everything from the previous session and continues exactly where it left off.

The Continuity Ledger

Section titled “The Continuity Ledger”

The ledger is a persistent notepad that survives context compaction. When the conversation history gets compressed to save tokens, the ledger retains the critical state needed to continue work.

What’s in the Ledger

Section titled “What’s in the Ledger”
## Session
**Started:** 2026-02-09 10:00
**Goal:** Refactor the authentication module to use JWT


## Recent Work
- [10:15] Created JWT utility functions
- [10:30] Updated login endpoint
- [10:45] Modified middleware to validate tokens


## Active Context
**Current focus:** Token validation middleware
**Files being modified:** auth/middleware.py, auth/jwt.py


## Key Decisions
- Using RS256 algorithm for JWT signing
- Token expiry set to 24 hours
- Refresh tokens stored in HttpOnly cookies


## Next Steps
1. Update integration tests
2. Deploy to staging

Managing the Ledger

Section titled “Managing the Ledger”
/ledger              # View current ledger
/ledger edit         # Edit manually

The AI updates the ledger automatically during complex tasks. You can also update it manually to leave notes for future sessions.

The .gee Directory

Section titled “The .gee Directory”

Every project has a .gee/ directory that stores all persistent state:

.gee/
├── gee.md                          # Project guidance (shared rules)
├── project.log                     # Cross-session activity log
├── activity.log                    # Detailed activity tracking
├── sessions/
│   └── {session-id}/
│       ├── continuity.md           # This session's ledger
│       └── session.log             # Session activity log
├── modes/                          # Mode-specific state
├── plans/                          # Saved execution plans
└── skills/                         # Project-local skills

Project Activity Log

Section titled “Project Activity Log”

Significant events are logged to .gee/project.log for cross-session visibility:

  • Commits (with message and files)
  • File modifications
  • Test results
  • Errors and resolutions

When starting a new session, Gee-Code reads recent project activity so the AI always knows what happened while you were away — even work done by other sessions or autonomous Gees.

Cross-Session Awareness

Section titled “Cross-Session Awareness”

When a session starts, the AI assembles context from:

SourceWhat It Provides
MemoryYour preferences, project knowledge, past decisions
Continuity ledgerCurrent work state, goals, blockers
Project activityWhat other sessions have done (commits, errors)
Project guidance.gee/gee.md — shared rules and conventions

This means the AI understands who you are, what you’re working on, what happened recently, and how this project works — all before you type your first message.

Multi-Session Coordination

Section titled “Multi-Session Coordination”

Multiple sessions can work on the same project simultaneously. The project log acts as a coordination mechanism:

  1. Session A commits code — logged to project.log
  2. Session B starts — reads activity — sees Session A’s commit
  3. Autonomous Gee activates — reads activity — knows what humans did

Each session has its own isolated ledger, so there are no conflicts.

Next Steps

Section titled “Next Steps”