StackMap
Subscribe

mini-swe-agent vs tau

The 100-line agent from the SWE-bench team: >74% on SWE-bench Verified with no tools but bash, no config sprawl — the reference minimal harness, adopted by Meta, NVIDIA and Ramp. — versus — Hugging Face's Python port of Pi's minimalist coding agent: a real terminal agent with TUI, sessions and skills — built to be READ, with a clean brain/environment/frontend separation.

The curated verdict

The two readable coding agents, at different altitudes: mini-swe-agent is the 100-line loop that proves harness complexity optional; Tau is the layered harness that teaches where the boundaries go when you DO add TUI, sessions and skills.

mini-swe-agenttau
Stars6.1k2.1k
Forks839226
LanguagePythonPython
LicenseMITMIT
Last activity4 days agotoday
Topicscoding, agentscoding, agents
Curated connections31

mini-swe-agent — the curator's take

The existence proof that most harness complexity is optional: the team that built SWE-bench and SWE-agent asked what a 100x simpler agent loses — the answer is almost nothing (>74% Verified), which is why it became the standard baseline harness for benchmarking models (Ramp's SWE-bench, DeepSWE — where it beats Claude Code and Codex as a harness). Read it to understand agents; use it to evaluate models fairly. NOT a daily driver: no MCP, no skills, no IDE plumbing — by design. If you're choosing a tool to ship features with, this is the control group, not the product.

tau — the curator's take

The teaching codebase for how coding agents are shaped: tau_ai (provider-neutral streaming) → tau_agent (the portable brain: loop, tools, events, sessions) → tau_coding (the actual app). That boundary discipline — the core knows nothing about rendering or config paths — is the lesson most production harnesses can't teach because they've already violated it. And unlike a toy, it's a usable daily agent with durable sessions. NOT the benchmark baseline (that's mini-swe-agent's 100 lines) and not the feature-race winner — no MCP sprawl by design. Read tau to understand the architecture; read mini to understand the loop; use whatever you like after.