DEDUPEDWho it is forA product lead choosing a harness and model for a feature and needing cost and speed data.
The problemThe lowest-cost and fastest harness setup changes by task, so guessing wastes money or adds latency.
The outcomeYou get a side-by-side view of how each harness and model pair handled the same task.
What you need first
- Docker installed
- About 4 GB of free disk space
- A provider API key for at least one supported model provider
- Python 3 if you use the Idempotency-Key example
Steps
- Complete setup and connect at least one provider under Bring Your Own Key
- Open Agent harnesses in the Console and choose a supported harness
- Select New task, pick a model, and give every configuration the same task text
- Review the live progress and the files each run produces
- Read the methodology at https://harnessrouter.ai/benchmarks before drawing conclusions
Prompt for your agent
Using the HarnessRouter repo at https://github.com/HarnessRouter/harnessrouter, run the same task <task-description> on two harness and model pairs in the Console. Expected output: a short table of which pair was faster and which produced the cheaper run. Verify that both runs used the same task text before reporting the comparison.
Caveat. The README says the fastest and lowest-cost configurations vary by task, so one run does not show which setup is best for other tasks.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for HarnessRouter/harnessrouter and "Compare harnesses on cost and latency for one task"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer