| Key features | - Works with Claude Code, Codex, Conductor, T3 Code
- Fast, Balanced and Thinking modes with automatic optimization
- Monthly budgets for organization, teams and individuals
- Alerts at 80% and 100% budget by email and Slack
- Spend tracking by repository, team and session
- Session history with steps, duration and cost
- Failover between providers when one slows down
- Audit log and device management
- Self-updating CLI for engineers
- Transcript control and compliance options
| - Configure plugins, skills, MCP servers, and hooks in a single harness.yaml file
- Observatory: monitor every session, tool call, and minute of activity
- Memory: persistent knowledge graph for AI orientation
- Comparator: run prompts across Claude Code, Cursor, and Copilot side by side
- Import existing configs from multiple AI coding tools automatically
- Share complete setup via single file with teammates
- Harness Parity: verify all connected AI tools end up with identical configuration
- Cross-harness compatibility table showing actual tool capabilities
|
|---|
| What makes it different | - Four times the work per dollar compared to Opus 5.5
- Engineers keep their existing tools, prompts and settings
- Intelligent cost optimization that picks thinking mode for each step
- Complete visibility into team AI spending and usage
- Engineers never hold provider keys, reducing security risk
| - Import-first approach: reads existing configs from all your tools and converges them before you write YAML
- No marketing gloss: capability table generated at build time from actual adapter code, not hand-maintained
- Portable across tools: single harness.yaml travels with you to every machine and tool
- Open source and free: no cost beyond existing Claude subscription
- Based on open specification: Harness Protocol is vendor-neutral and JSON Schema validated
|
|---|