DOCUMENTATIONAEGIS

Everything you need
to ship reliable agents.

Guides, API reference, SDK docs, and self-hosting instructions. Start in 5 minutes or go deep on the architecture.

Install

npm install @aegis/sdk
# or
pip install aegis-sdk

Initialize

import { Aegis } from '@aegis/sdk'

const aegis = new Aegis({
  apiKey: process.env.AEGIS_API_KEY,
})

Run evals

const result = await aegis.eval.run({
  suite: 'production-v2',
  agent: myAgentFn,
  sandbox: true,
})

console.log(result.passRate) // 97.4
All sections

Getting Started

Install the SDK, connect your repo, and run your first eval in under 5 minutes.

Eval Stack

Deep dives into sandbox isolation, LLM proxy, eval engine, telemetry, and transcripts.

CI Integration

Connect Aegis to GitHub Actions, GitLab CI, and any other CI provider.

API Reference

Full REST API reference with request/response schemas and code examples.

SDK Reference

TypeScript and Python SDK documentation, types, and advanced usage patterns.

Self-hosting

Deploy Aegis on your own infrastructure — Docker, Kubernetes, bare metal.

Architecture

How Aegis fits into
your stack.

Aegis sits between your agent code and your CI pipeline. It wraps each eval run in an isolated sandbox, routes LLM calls through its proxy, runs your eval suite, and reports results back to your PR as a status check.

01Your agent code calls the Aegis SDK
02Aegis spins up a Firecracker microVM sandbox
03All LLM calls route through the Aegis proxy
04Eval suite runs and scores responses
05Pass/fail result posted to your PR