Tau coding agent
SN66Autonomous AI agents that manage servers, deploy code, and handle DevOps for you
A staged software engineering evaluation workflow. Mining tasks are generated from real GitHub commits, solver agents compete to produce code fixes, and results are scored by both changed-line similarity and LLM-based judging. The best agent earns the most emissions.
// Benchmarking coding agents by competition.
Tau is a CLI-based evaluation framework for coding agents. The on-chain identity describes it as a "coding agent" focused on "distilling software agents." The GitHub repository at github.com/unarbos/tau implements a staged workflow where agents are tested on real software engineering tasks.
The simple version: Take a real bug from a real open-source project. Give it to 10 different AI coding agents. See which one actually fixes it correctly. Tau is the system that runs that tournament.
How it works:
generatemines a commit from GitHub and creates a coding tasksolveruns a solver agent against that task (supports Cursor CLI, Claude CLI, Docker-sandboxed agents, or any agent hosted on GitHub)comparescores two solutions by changed-line similarityevalcompares multiple solutions using an LLM judgedeleteremoves saved artifacts
Solvers run in Docker containers with resource limits. Evaluation uses both line-level diff comparison and LLM-based judging.
- The problem it solves: AI coding agents are proliferating but it's hard to compare them objectively. Running the same benchmark on different agents and scoring them fairly requires infrastructure.
- The opportunity: The AI coding assistant market is growing rapidly. A decentralized evaluation framework that surfaces the best agent creates continuous quality improvement pressure.
- Traction signals: 27 commits in the last 4 weeks across 2 contributors (unarbos, unconst). 43,136 TAO market cap. Momentum score 70 (the highest among subnets covered in this batch). Net 7-day inflow of 1,108 TAO. Gini 0.585 (well-distributed). 100% miner burn with 0 active miners currently.
Category: Code Generation and Development Tools
Tau is notable for two reasons: the builder and the pace.
Mechanism:
The staged workflow is designed for reproducible evaluation. Generate creates a task from a real GitHub commit. Solve runs agents against it. Compare and eval score the results. This means the same coding challenge can be benchmarked across multiple agents with consistent scoring.
The solver backends are flexible: Cursor CLI (in Docker), Claude CLI (on host), OpenRouter-powered Docker agents, or any agent hosted on a GitHub repo. The bundled agent lives in tau/agent and can be customized or shared via GitHub.
The "king" mechanic referenced in recent commits ("auto-update king commitment when chain data changes", "resolve king SHA from state") suggests a king-of-the-hill weight system similar to other subnets by the same builder.
The repo was created March 31, 2026, and has 27 commits in the last 4 weeks across 2 contributors. This is active development.
Market metrics show strong momentum. Momentum score of 70 is the highest among recent subnet coverage. Net 7-day inflow of 1,108 TAO is one of the largest positive flows. The 90-day return of 123% and 30-day return of 118% indicate strong market interest.
- Zero active miners: The subnet is not currently distributing emissions to miners. 100% burn indicates a burn or transition period.
- Two contributors: Small team, though both (unarbos and unconst) are known Bittensor builders.
- Development is very recent: Repo created March 31. The tool is new and unproven at scale.
