What It Is
The Forge Genome is a persistent knowledge structure that accumulates model behavior data across every session. It tracks which models excel at which tasks, which prompting strategies produce better results, tool usage patterns, failure modes, and recovery strategies. Unlike a chat history, the genome is structured intelligence — not raw conversation logs.
How It Works
After every session (and periodically every 25 turns), Forge collects a genome snapshot: model performance metrics, tool call success rates, context swap patterns, behavioral fingerprint data, and break/assurance results. This snapshot is merged into the persistent genome file, building a longitudinal profile of model behavior on your specific workload.
Fleet Sync
Pro and Power tiers unlock genome sync across fleet installations. When one team member discovers that a particular model handles TypeScript refactoring better than Python, that intelligence propagates to the entire fleet. The AI gets smarter for everyone as anyone uses it. Sync happens automatically via the master/puppet architecture — no manual configuration.
What It Tracks
- Model affinity — which models perform best for which task categories on your codebase
- Tool patterns — which tools are used most effectively, failure rates per tool
- Context efficiency — how many turns before context degradation, optimal swap thresholds
- Break/Assurance history — reliability trends over time, regression detection
- Behavioral drift — fingerprint changes across model versions
- Recovery strategies — which AMI recovery actions work for which failure modes
Why It Matters
Most AI tools treat every session as a blank slate. Forge remembers. A model that failed at SQL optimization last week won't be routed SQL tasks this week (if router is enabled). A tool that consistently times out gets deprioritized. The genome turns Forge from a stateless wrapper into a system that learns and adapts to your specific development environment.