omp, short for oh-my-pi, is an MIT-licensed terminal coding agent forked from Pi, with LSP, debugger and browser tools built in. HagiCode is a desktop and web workbench that runs omp as its executor, so the two work together rather than compete.
omp is oh-my-pi, a terminal coding agent you install and run from the command line. It started as a fork of Pi and adds IDE-style tooling such as LSP, a debugger, subagents and a browser. HagiCode is a different layer: a desktop and web workbench that launches agent CLIs, including omp, and gives you sessions, specs, commits and history around them. It does not replace omp, and it does not host models.
If you work alone in a terminal, run omp interactively or in scripts, and are happy tracking your own tasks, stay with omp. Add HagiCode when you want several omp tasks running in parallel, with a written proposal and a reviewable history for each one.
| Aspect | omp (oh-my-pi) | HagiCode |
|---|---|---|
| What it is | Terminal coding agent, a fork of Pi, with LSP, debugger, browser and subagent tools built in | AI programming workbench that runs agent CLIs (including omp) as executors |
| Interface and platforms | Terminal UI, one-shot print mode, RPC, SDK, ACP for editors; macOS, Linux, Windows | Desktop app (Windows, macOS, Linux), Docker Compose container, Microsoft Store edition |
| Pricing and licensing | MIT-licensed and free; you pay only for the models you connect | Desktop and Container editions free; Store edition free base, optional one-time TurboEngine purchase |
| Model and provider access | Many providers (README says 60+), custom OpenAI-compatible endpoints, per-role model routing | Hosts no models; uses omp's own sign-in, or OmniRoute routing if you configure it |
| Parallel work | Subagents run in isolated worktrees inside one session | Several sessions side by side, managed as named Agents |
| Spec and planning workflow | Planning model role and a review command; proposal tracking is up to you | OpenSpec proposals: propose, validate, apply, archive |
| Git and commit handling | Built-in omp commit splits changes into atomic commits | AI-written commit messages with repository locking; Git management in the UI |
| Asking you questions | Built-in structured questions in its terminal UI | Shows omp's questions in the conversation, then continues the same run |
| Setup effort | One install command, then sign in or add an API key | Install omp first, then HagiCode, then pick OMP as the Agent CLI |
| Best for | Terminal-first developers who want deep tooling and scripting | Running several omp tasks across repos with specs, review and history in one place |
The README lists 14 LSP operations and 28 DAP operations. File renames go through the language server, so imports update before a file moves. Debugger sessions work with lldb-dap, dlv and debugpy.
It also keeps persistent Python and Bun workers that can call the agent's own tools, and drives Chromium tabs for browser work.
Hashline edits anchor each change to a content hash and are rejected if the file has moved on. ast_edit previews a structural change before it is applied.
Stream rules let a regex match abort the stream and inject a rule before the model retries, which is useful for keeping a model on your conventions.
The task tool fans work out to subagents in isolated worktrees and returns schema-validated results. An advisor role lets a second model review each turn.
Models are routed by role (default, smol, slow, plan, commit, vision, task, advisor, tiny), with fallback chains and path-scoped model sets configured in ~/.omp/agent/config.yml.
You can run it as a terminal UI, a one-shot omp -p command, an RPC process over stdio, a Node SDK, or an ACP server for editors such as Zed.
It reads rules and MCP servers from .claude, .cursor, .codex, .cline and other tool folders, so moving to omp does not mean rewriting your project instructions.
HagiCode does not bundle or replace omp. It looks for the omp executable on your machine, checks its version, and starts it for each run, so your omp install, sign-in and configuration stay exactly as they are.
HagiCode passes the working directory and the model you chose, and can set omp's provider, thinking level, no-session mode and tool restrictions from its settings.
Each omp run in HagiCode is its own session, so you can have one working on the API while another writes tests. Agents management gives every session a name and a visible status instead of a row of terminal tabs.
A task starts as a proposal and moves through propose, validate, apply and archive. The reasoning behind an omp change stays next to the diff. Free editions allow 3 concurrent proposals.
HagiCode writes the commit message from your diff and locks the repository while it commits. The embedded Code Server opens the file omp changed in a browser editor. omp's own commit command still works in a terminal.
Monospecs declares your repositories in one file so omp starts a proposal with the full change surface. Vault stores reference projects and notes that new proposals can use as context.
If you configure OmniRoute, omp's model requests can be routed through it. Tool calls from omp are shown in the same format as other agents. An optional status-bar item shows today's tokens and cost from TokenMonitor, off by default.
When omp asks a structured question, HagiCode shows it in the conversation, the session waits for a reply, and your answer continues the same run. If the run already ended, the answer goes in as a normal message in the same session.
HagiCode is an agentic coding workspace. Rich, convenient UI interactions make planning, agent execution, and project work approachable and easy to navigate, creating an engaging development experience.
Explore an engaging development experience through rich, convenient UI interactions.
HagiCode is an agentic coding workspace. OpenSpec workflows plan the change, multi-agent execution carries it out, and Hero Scenario makes the long run visible, so ideas become shipped software.
© 2026 HagiCode | Powered By hagilight-starlight@0.20.0 | Branch: main | Commit: 0dde182 | Built:
