Empryo Integration

Empryo is a graph-powered AI coding agent — it parses your codebase into a code-intelligence graph and edits symbols, not strings. It runs as a terminal UI, a desktop app, and a headless CLI on macOS, Linux, and Windows, and LLM Gateway is built in: one browser login (or one pasted key) connects all ~190 text models on the gateway. Every request is tagged with x-source: empryo, so usage is attributed correctly in your dashboard and on your DevPass plan.

Install

macOS / Linux:

curl -fsSL https://empryo.com/install.sh | bash

Windows (PowerShell):

irm https://empryo.com/install.ps1 | iex

Each script downloads the release's SHA256SUMS.txt, checks its RSA-4096 signature against a key pinned in the script, and verifies the binary's checksum before installing it — the one-liner itself is a bootstrapper, so if you'd rather not pipe it into your shell, download it first and read it, or grab a binary and its checksum from the download page and verify by hand.

Setup

empryo

One connection, three surfaces

The gateway connection is shared by everything Empryo ships:

• Terminal UI — /login, /keys, and /models as above.
• Desktop app — the same browser login (from onboarding or the keys drawer) and the same grouped model picker.
• Headless CLI — for scripts and CI:

empryo --headless "fix the failing test" --model llmgateway/claude-opus-5

Add --json for a structured result or --events for a real-time JSONL stream.

Model routing

Empryo passes model IDs to the gateway verbatim, so both forms work:

• Canonical (claude-opus-5) — the gateway routes to the best available provider, with automatic failover. Required on DevPass plans.
• Provider-pinned (anthropic/claude-opus-5) — pay-as-you-go keys can pin a specific upstream deployment. See the routing docs for the full syntax.

Claude models take the gateway's Anthropic-native endpoint (/v1/messages) rather than the OpenAI-compatible one, so extended thinking and prompt caching (cache_control) survive the hop. Other models ride /v1/chat/completions, with reasoning effort injected where the model supports it.

DevPass meters, natively

With a DevPass key connected, Empryo reads the gateway's GET /v1/key endpoint and renders your plan meters in the product: monthly credits and the weekly premium fair-use window show up as a footer badge in the TUI and in the desktop usage drawer, next to Empryo's own usage ledger (/usage), which tracks what every session cost locally.

Why use LLM Gateway with Empryo

• ~190 text models — Claude, GPT, Gemini, Grok, DeepSeek, and more through one login
• Live pricing in the picker — rates, context, and output caps come from the gateway's catalog, not a stale copy
• Native Anthropic wire for Claude — thinking and prompt caching intact
• Attribution built in — every request carries x-source: empryo, so the dashboard shows exactly what the agent spent
• DevPass aware — plan meters render inside Empryo, and billing is handled by the gateway

Troubleshooting

"LLM Gateway is set to use the browser login, but you're not connected"

Run /login, or switch to a pasted key with /keys (and set llmgatewayAuth: "key" if you want the key to win permanently).

"Direct provider routing is not available on coding plans"

Your key belongs to a DevPass plan and you picked a provider-pinned ID. Choose the canonical form of the model instead — claude-opus-5, not anthropic/claude-opus-5.

Models show no price

The catalog is fetched live from the gateway at startup. Check your connection and that your key is valid in the dashboard, then restart Empryo.