Empryo vs pi coding agent
Compare Empryo’s repo map, desktop app and mobile companions with a harness you can extend or embed through a TypeScript SDK. Here, pi means the coding agent in earendil-works/pi, previously badlogic/pi-mono. Empryo provides AST edits for TypeScript and JavaScript, plus navigation and renames through supported language servers. Coverage depends on the language and project configuration.
A persistent repo map, or a harness you extend yourself.
6 rows: 6 different
Side by side
| Feature | Empryo | pi coding agent | In short |
|---|---|---|---|
| Starting point | A built-in ranked map of repository files, symbols and imports. | A minimal extensible harness, customized with extensions and packages. | Different approach Choose how much of the workflow you want supplied or assembled. |
| Code changes | Empryo provides AST edits for TypeScript and JavaScript, plus navigation and renames through supported language servers. Coverage depends on the language and project configuration. | A coding-agent workflow that you can adapt with extensions. | Different approach Test a cross-file change; inspect references and the resulting diff. |
| Interfaces | macOS, Windows and Linux desktop/terminal apps, headless CI, and beta iPhone, iPad, Apple Watch and Android companions. The engine runs on your computer. | Interactive terminal, print and JSON modes, RPC and a TypeScript SDK. | Different approach Empryo supplies a desktop UI; pi documents APIs for building integrations. |
| Customization | Skills, MCP servers and desktop Morphs. | Extensions, skills, prompt templates, themes and distributable pi packages. | Different approach Start with the extension you actually need. |
| Models | Built-in providers, custom compatible endpoints and local model providers. | Connect a built-in provider through /login using a subscription or API key. | Different approach Verify your intended provider and account before comparing cost. |
| Evidence | Published small-sample runs against historical pi versions, including provider-bill checks. | Those historical results do not describe every current pi release or extension setup. | Different approach Use the original run dates and settings; do not infer a current winner. |
- Different approach
Build your own workspace and agent workflows
Morph the app around your work
Composer · panels · full canvas
Build a board, notebook or control panel from one JSON file. Add commands and app extensions, restyle the composer and chat frame, or fill the workspace with an interactive canvas while chats keep running behind it.
Build your own workspaceRevise plans while work continues
Review · approve · steer · revise
Inspect steps, files, diffs and verification criteria. Add points or comments to a step while the agent works; feedback reaches its steering queue. Review revisions, keep drafts and follow progress without starting a new chat.
Revise a plan during executionEngineer agent graphs
Planner · parallel reviewers · judge · retry
Connect agents, shell steps, councils and approval gates in a workflow graph. Branch on results, bound retries and assign models per role. Inspect the live run graph alongside the separate Genome map of your code.
Build a review-gated workflowOrganize projects and tab teams
Projects · worktrees · child tabs · links
Keep multiple projects in one workspace. Give each tab its own model and history, open nested child tabs, exchange messages and hand off work. File claims warn about overlapping edits; Git protections block destructive cross-tab operations.
Child tabs · Coordinate a tab teamAsk for another perspective
Duck · advisor · council · reviewer
Think aloud with a duck, consult a separate model, convene a debating council or ask an independent reviewer to inspect a change. Choose models and effort for each role, and keep the findings alongside the work.
Advisors and councils · ReviewInspect what the engine is doing
Console · model calls · costs · compaction
Open the console and workbench to inspect errors, model requests, tokens, latency and compactions. Follow background agents and graph runs, inspect changes and checkpoints, and use embedded terminals rather than switching applications.
Console and logs · WorkbenchOne engine, across your devices
Desktop, terminal and CI
Work on macOS, Windows or Linux. Use the desktop workspace, terminal UI or headless CLI with the same engine, tools and sessions.
Explore the interfacesPhone, tablet and watch
Beta companions for iPhone, iPad, Apple Watch and Android let you steer chats, answer approvals, switch projects and send voice prompts. The agent runs on your computer.
Continue from your phoneLAN, Tailscale or relay
Connect on your local network, through Tailscale, or through the internet relay without a Tailscale account. Pair with approval; optional background access keeps the host available after the app closes. The host must remain powered and reachable.
Remote setupPublic core, private additions
The v1/v2 core and engine code are public. Newer v3 features and desktop/mobile surfaces are developed privately. The public code retains its existing license; v3 does not revoke those rights.
Read the public coreThe public repository uses Business Source License 1.1, a source-available license with an Apache 2.0 change date. Read its terms.
Models and parallel agents
Choose providers and local models. Assign models and effort to research, coding and review, or run specialist agents in parallel with approval and retry steps.
Model routing · Multi-agent workflowsMemory and recovery
Keep project decisions and preferences across sessions. Inspect checkpoints, rewind files and conversation, or take another approach from an earlier turn.
Project memory · CheckpointsWhich one fits
Choose Empryo if
You want the agent to start with a ranked map of files and dependencies, use AST tools for TypeScript and JavaScript, and work from a desktop app or terminal without building that interface yourself.
Choose pi coding agent if
You want to shape the harness with extensions, skills, prompt templates and themes, or build an application around its SDK. Pi’s documentation leaves features such as subagents and plan mode to customization or packages.
Bottom line
The choice is how you want to work: Empryo supplies a repo map and desktop workflow; pi exposes a harness you can adapt and embed. Run the same acceptance test in both before choosing.
Sources and evidence
Documentation checked . This is a workflow comparison, not a benchmark ranking.
Common questions
Which pi does this page mean?
The pi coding agent in earendil-works/pi, previously hosted at badlogic/pi-mono. It is not the Pi conversational assistant or Raspberry Pi.
Do the published benchmarks prove Empryo wins today?
No. They cover particular tasks, versions and models. Read the original methodology and reproduce the task with the versions you plan to use.