Developer Docs

Introduction

AmirCrew Labs provides autonomous agent infrastructure for web3 institutions, research teams, and developers. Agents connect to external systems through MCP servers, reason with the Claude API, and execute tasks inside isolated Docker sandboxes, 24/7, without human babysitting.

Three capabilities define the platform:

• Autonomous Agentic Gateway bridges agents to messaging platforms and automation systems, so chat becomes a live command surface.
• On-Chain & Document Synthesizer ingests thick whitepapers, DAO proposals, and raw transaction data into structured signal using a massive context window.
• Enterprise-Grade Security & Secrets covers credential vaulting, API key management, and encryption for continuous operation.

Quickstart

Get an autonomous agent running locally in three steps.

1. Install the CLI

npm i -g @amircrew/cli

2. Authenticate

amircrew login

The CLI opens a browser session and stores a scoped token in your OS keychain, never in plaintext.

3. Run your first agent

# create a project + scaffold an agent
amircrew init my-agent
cd my-agent

# start the agent runtime
amircrew run

→ booting autonomous agent ac-01 …  ready
→ mcp servers: onchain, documents, telegram, secrets
✓ agent listening on http://localhost:8787

Installation

Requirements: Node.js 20+ or Python 3.11+, and Docker for sandboxed execution.

Node.js

npm i -g @amircrew/cli

Python

pip install amircrew

Docker (recommended)

docker pull ghcr.io/amircrew/runtime:latest

Authentication

All API requests authenticate with a bearer token. Create a key from the console, then export it:

export AMIRCREW_API_KEY="ac_live_••••••••••••••••"

Send it on every request:

curl https://api.amircrew.org/v1/agents \
  -H "Authorization: Bearer $AMIRCREW_API_KEY"

Autonomous Agents

An agent is a long-running process that reasons with Claude, calls tools exposed by MCP servers, and acts on results. Agents are declarative: you describe goals and tools, not step-by-step scripts.

import { Agent } from "@amircrew/sdk";

const agent = new Agent({
  name: "treasury-monitor",
  model: "claude-sonnet",
  tools: ["onchain.read", "documents.search", "telegram.send"],
  schedule: "*/15 * * * *", // every 15 min
});

agent.on("anomaly", async (e) => {
  await agent.tools.telegram.send(`⚠️ ${e.summary}`);
});

Model Context Protocol

MCP is the open standard that connects agents to tools and data. AmirCrew ships four first-party servers, and you can register your own.

Registering a custom server

amircrew mcp add my-server --command "node ./server.js"

Agentic Gateway

The gateway routes inbound messages from connected platforms to the right agent, maintains conversation context, and streams responses back. Configure a route in amircrew.config.json:

{
  "gateway": {
    "routes": [
      { "source": "telegram", "agent": "treasury-monitor" }
    ]
  }
}

On-Chain & Document Synthesizer

Ingest large corpora and transaction histories, then ask questions in natural language. The synthesizer keeps a massive context window so it never loses the thread across long documents.

const result = await agent.documents.synthesize({
  sources: ["proposal-412.pdf", "treasury-2025.csv"],
  question: "What are the top 3 risks and their on-chain evidence?",
  citations: true
});

console.log(result.summary);

Docker Sandbox

Every agent task that executes code runs inside an ephemeral container with a read-only filesystem, no network by default, and CPU/memory limits. Enable network access per task:

await agent.exec({
  cmd: "python analyze.py",
  sandbox: { network: false, cpu: "1", memory: "512Mi", timeout: 30000 }
});

REST API

Base URL: https://api.amircrew.org/v1. All endpoints require a bearer token.

Example: create an agent

curl -X POST https://api.amircrew.org/v1/agents \
  -H "Authorization: Bearer $AMIRCREW_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name":"treasury-monitor","model":"claude-sonnet"}'

{
  "id": "agt_9f2c1b",
  "name": "treasury-monitor",
  "status": "active",
  "created_at": "2026-01-14T09:12:03Z"
}

CLI Commands

Security & Secrets

Credentials are encrypted at rest with envelope encryption and decrypted only inside the agent runtime. Secrets are never written to logs or exposed to sandboxed code unless explicitly injected.

• Envelope encryption: per-project data keys wrapped by a root key.
• Scoped tokens: keys are limited to specific agents and actions.
• Automatic rotation: rotate keys on a schedule with amircrew secrets rotate.
• Audit trail: every secret access is logged with actor, time, and agent id.

Error Codes

Telegram Gateway

Connect an agent to Telegram in two steps.

1. Link your bot

amircrew gateway connect telegram --token "$BOT_TOKEN"

2. Route messages to an agent

amircrew gateway route telegram --agent treasury-monitor --chat "@my_channel"

Messages from that chat now reach the agent, and replies stream back to Telegram.

Deployment

Deploy an agent runtime to AmirCrew Cloud with a single command:

amircrew deploy --prod

Or self-host with Docker:

docker run -d --name ac-runtime \
  -e AMIRCREW_API_KEY=$AMIRCREW_API_KEY \
  -p 8787:8787 \
  ghcr.io/amircrew/runtime:latest

FAQ

Which model does AmirCrew use?

Agents run on the Claude API by default; the model is configurable per agent.

Do I need Docker?

Docker is required for full sandbox isolation. Without it, code runs in-process with reduced isolation.

Can I bring my own MCP servers?

Yes. Register any MCP-compatible server with amircrew mcp add.

Is there a rate limit?

Default limits are per-plan; exceeded limits return 429. Contact us for enterprise quotas.