Code has structure. Empryo maps symbols, callers and dependencies, then edits code by symbol instead of relying only on text replacement. Review diffs and plans, keep project decisions across sessions, and choose a model for each job. Desktop GUI, terminal TUI and headless CLI on macOS, Linux and Windows.

hey PH! i built Empryo around one idea: code has structure, and the agent should use it. Functions have callers. Symbols have scope. Files depend on each other. I wanted the agent to follow those relationships and edit by name, instead of rebuilding that understanding through repeated searches. You can review the diff, steer a plan while it runs, keep project decisions across sessions and choose a different model for each job. The same engine runs on desktop, in the terminal and headlessly. I also measured the tradeoff. In the August Opus 5 benchmark, Empryo used 37 tools against pi's four. Both fixed all four bugs; the round cost $2.27 with Empryo and $3.10 with pi. pi was cheaper on the simplest task. The setup and per-task results are at https://empryo.com/benchmarks/fo... The demo shows the workflow. I'd like to hear where it helps on your repo and where it still gets in the way. The little guy is Mote.

I've been using Empryo while building LLM Gateway, DevPass, Lounge and Airside, for code changes, documentation, landing pages and launch assets. Keeping project decisions across sessions is especially useful when switching between all four.

One concrete number: my September usage logs show around 404 million cached input tokens, worth about $2,056 in estimated prompt-cache savings versus uncached API pricing. Most of my usage is on subscriptions, so that's API-equivalent token value, not cash taken off my bill.

Congrats on the launch, Ouail!

Empryo

The coding agent that knows your code by heart

The coding agent that knows your code by heart