oh-my-pi: a terminal agent that treats the harness as the product

June 26, 2026·AI Newsroom·4 min read·11 sources

oh-my-pi omp ai-coding-agent harness harness-engineering hashline+24high-risk claims

Screenshot of the oh-my-pi GitHub README showing the 'Every tool, benchmaxed' table with four model rows (Grok Code Fast 1 6.7% to 68.3%, Gemini 3 Flash +5 pp, Grok 4 Fast minus 61 percent tokens, MiniMax 2.1x), the 'The Pi you love, with batteries included' heading, the '01 Code execution w/ tool-calling' section intro, and the fork provenance line 'Originally built on Mario Zechner's wonderful Pi, omp adds everything you're missing.' — Source: github.com/can1357/oh-my-pi/blob/main/README.md · Captured 2026-06-26 via Playwright Chromium · License: MIT (project) / screenshot used editorially for the oh-my-pi article

On 2026-06-26, can1357/oh-my-pi — a fork of Mario Zechner’s pi-mono, re-engineered into a terminal AI coding agent called omp — sat at 14,677 stars, 1,287 forks, 369 open issues, 10,671 commits, MIT, created 2025-12-31 (metadata). Three releases (v16.1.19, v16.1.20, v16.1.21) shipped in the 18 hours before publication. That cadence is the signal worth understanding.

What it is

A terminal coding agent, not a wrapper. omp ships as a single Rust binary on macOS, Linux, Windows (no WSL bridge). Install: curl -fsSL https://omp.sh/install | sh on Unix, irm https://omp.sh/install.ps1 | iex on Windows, brew install can1357/tap/omp, or bun install -g @oh-my-pi/pi-coding-agent. Four entry points: TUI, one-shot prompt, Node SDK, stdio RPC / ACP.

The fork relationship. oh-my-pi is a fork of Mario Zechner’s Pi — previously at badlogic/pi-mono. LICENSE: ”© 2025 Mario Zechner / © 2025-2026 Can Bölük.” Upstream Pi: 65.7k stars, MIT, 240 releases. The maintainer’s framing: “The model is the moat. The harness is the bridge. Burning bridges just means fewer people bother to cross.” (blog.can.ac, 2026-02-12).

The four numbers. 40+ providers (Anthropic, OpenAI, Gemini, xAI Grok, Mistral, Groq, Cerebras, Fireworks, Together, Hugging Face, NVIDIA NIM, OpenRouter, Ollama, llama.cpp, vLLM, LiteLLM). 32 built-in tools plus 14 LSP + 28 DAP operations. ~55k lines of Rust in the N-API addon.

Why it matters

The harness is the boundary between “the model knows what to change” and “the change is on disk.” Most agents leave that boundary to a thin wrapper around read, str_replace, and bash. omp treats it as a product surface. Four mechanisms carry the load.

1 · Hashline — edit by content hash. The harness tags every read line with a 2-3 hex character content hash; every patch is anchored to a [PATH#TAG] full-file content hash, and rejects the patch before it corrupts anything if the file has changed. The model never retypes the lines it wants to change; the harness resolves the anchor and applies SWAP / INS / DEL operations (hashline README).

2 · First-class subagents with schema-validated yield. The task tool fans out into isolated worktrees; the final yield is a schema-validated object the parent reads directly — no prose to parse.

3 · Time-traveling stream rules. A regex match against the streaming model output aborts mid-token, injects the rule as a system reminder, and retries from the same point. README example: the model is about to write Box::leak; the request aborts, an amber card injects “Don’t reach for Box::leak in production code paths,” and the agent course-corrects to Arc<str>.

4 · The advisor model. A second model on the advisor role reads every turn and injects notes inline. Stream rules catch mechanical departures from policy; the advisor catches semantic drift.

Plus: native on Windows (ripgrep, glob, find, brush in-process); eight-format config import (Cursor MDC, Cline .clinerules, Codex AGENTS.md, Copilot applyTo); 18-hour release cadence with fixes at the right depth.

The benchmark

The README publishes a 540-task, 16-model benchmark (3 runs per task, fresh session each time):

Model	Metric	What
Grok Code Fast 1	6.7% → 68.3%	Tenfold lift when the edit format stops eating the model
Gemini 3 Flash	+5 pp	Over `str_replace` — beats Google’s own best attempt
Grok 4 Fast	−61% tokens	Output collapses once the retry loop on bad diffs disappears
MiniMax	2.1×	Pass rate more than doubles, same weights, same prompt

All four are project-published and self-attributed. Treat the spread as the order-of-magnitude the harness contribution can deliver, not a contract (README).

Practical implications

For harness evaluators. The right question is “which agent’s tool boundary is best” — not “which agent is best.” All of omp’s load-bearing moves are harness-side.
For teams replacing Claude Code / Codex CLI today. Migration cost is low — omp reads Cursor MDC, Cline .clinerules, Codex AGENTS.md, Copilot applyTo in their native shape, and the same binary runs against Anthropic, OpenAI, xAI, Google, Mistral, Groq, Ollama, or llama.cpp.
For researchers. Hashline lives in packages/hashline; the agent runtime in packages/agent exposes typed events.

What to watch

Independent reproduction of the 16-model × 3-edit-format matrix on a model not in the original sixteen.
Upstream Pi merge cadence — omp is a fork, not a vendor branch.
The advisor model in real sessions — no published benchmark yet.
/collab and the my.omp.sh relay — third-party service, no SLO.
VOUCHED.td governance — a public denouncement list of bad actors.

Risks and caveats

The 10× lift numbers are project-published and self-attributed. Treat as the order-of-magnitude the harness contribution can deliver, not a bill-line guarantee.
The my.omp.sh relay is a third-party dependency. Frames are sealed client-side; that protects content, not availability.
Supply-chain surface is large. 40+ provider SDKs, 14 web-search backends, MCP, ACP, Discord, four platform-tagged binaries.
369 open issues / 10,671 commits is a velocity signal, not a stability signal. Six tags in 72 hours.
MIT is permissive; the maintenance commitment is not guaranteed. Pin a release tag. Vouch-based contribution model; one maintainer’s call.

Verdict

can1357/oh-my-pi — the fork of Mario Zechner’s Pi shipping as omp — sat at 14,677 stars, 1,287 forks, 369 open issues, 10,671 commits, MIT, with a Rust core of ~55k lines, 40+ LLM providers, and a project-published 16-model edit benchmark. The 18-hour release cadence is the second load-bearing signal: a project using its own binary. The takeaway is harness engineering — the model is the moat, the harness is the bridge — a working MIT-licensed demo of a serious agent boundary in 2026. Pin a release tag, read packages/hashline.