A lightweight multi-agent AI systems. Stop rebuilding orchestrators, database agents, and retry logic. Just define your workflow in YAML, customize with your data, and ship.
Building multi-agent AI systems means rebuilding the same things over and over:
- Orchestrators that route tasks
- Database agents with SQL guardrails
- Retry logic and backoff strategies
- Loop limiting (preventing rogue agents)
- Cost tracking and observability
Most frameworks are bulky and complex. You just want pre-built components you can compose—like AWS services.
Echos lets you:
- Coordinate multiple AI agents - Let different agents handle different tasks
- See what's happening - Watch your agents work in real-time
- Debug easily - Visual traces show you exactly what went wrong
Echos gives you pre-built agent "services" you compose like AWS:
| What You Need | Echos Agent | What It Does |
|---|---|---|
| Query databases | db_agent |
SQL with guardrails (no DELETE/DROP) |
| Call HTTP APIs | api_agent |
SSRF protection, domain whitelisting |
| Search the web | search_agent |
Serper/Brave integration |
| Analyze data | data_agent |
Summaries, insights, charts |
| Generate code | code_agent |
Multi-language code generation |
| Route tasks | orchestrator |
Intelligent task routing |
Define your "architecture" in YAML (not code):
agents:
- name: orchestrator
maxLoops: 3
- name: db_agent
policy:
guardrails:
allowedTables: [users, orders]
allowedOperations: [SELECT]
retries:
count: 2
backoffMs: 300
routes:
orchestrator:
canCall: [db_agent, data_agent]Try the product live at https://echoshq.com
Join our Discord community for discussions, support, and updates:
Echos is local-first and designed for sensitive data:
- Your workflow data never leaves your machine - All execution happens locally
- Runs on localhost - No external dependencies required
- API keys authenticate to YOUR local server - Not an external service (localhost:4000)
- Optional external calls - Only LLM APIs (OpenAI, Anthropic, AWS Bedrock) go external. Use AWS Bedrock in your VPC or local models (Ollama) for complete data isolation
- Multi-tenant ready - Organization isolation built-in
For Production:
- Self-host in your VPC or deploy to any infrastructure.
- Use it at Echos
Use anywhere:
import { EchosRuntime } from '@echoshq/runtime';
const runtime = new EchosRuntime({
apiKey: process.env.ECHOS_API_KEY,
apiUrl: process.env.ECHOS_API_URL || 'http://localhost:4000'
workflow: './workflow.yaml' // optional, defaults to 'workflow.yaml'
});
// Simple usage
await runtime.run('Analyze Q4 sales by region');
// Or with memory context
await runtime.run({
task: 'Analyze Q4 sales by region',
memory: { year: 2024, region: 'north' }
});When an agent fails in production, you can now:
- Click the failed trace
- Click "Time Travel Debug"
- Edit the workflow config right there
- See what WOULD have happened with your changes
- Deploy the fix instantly
You just fixed 47 production failures in 5 minutes.
This is like Chrome DevTools time-travel debugging, but for AI agents.
Paste any OpenAPI spec → instant production-ready agent.
Click "Generate from OpenAPI" in the UI, paste a Stripe, GitHub, or any API spec, and get:
- Auto-configured guardrails - Domain allowlisting, SSRF protection
- All endpoints mapped - Complete API documentation in agent memory
- Auth detection - Bearer, API Key, OAuth2 automatically configured
- Instant deployment - Preview, customize, and save in seconds
Example:
# Paste this URL:
https://raw.githubusercontent.com/stripe/openapi/master/openapi/spec3.json
# Get a Stripe agent that can:
✓ Create customers
✓ Process payments
✓ Manage subscriptions
✓ Handle refunds
# All with natural language!Turn any API into a natural language interface in under 30 seconds.
Join our Discord community for discussions, support, and updates:
- Time-Travel Debugging - Click any failed trace → Edit config → Test the fix → Deploy. Fix production issues in 60 seconds.
- Cost Estimation & Warnings - See estimated costs before running workflows. Get warned about high-cost operations (>$0.10)
- Loop Limiting - Set
maxLoopsper agent to prevent infinite loops - Retry Logic - Automatic retries with exponential backoff
- Cost Ceilings - Per-agent and per-workflow cost limits
- Memory Management - Agents share context through namespaced memory
- Multi-LLM - OpenAI (GPT-3.5, GPT-4, GPT-4o), Anthropic (Claude), or AWS Bedrock (Claude, Titan, Llama, etc.)
- Visual Traces - See what happened, where it failed, and costs
- Guardrails - SQL injection protection, SSRF blocking, table/domain whitelisting
- Environment Validation - Automatic checks for Docker, Node.js, and dependencies on startup
Everything runs on your machine—no external services required.
# 1. Start Echos locally (validates environment, starts PostgreSQL + server + web UI)
git clone https://github.com/treadiehq/echos.git
cd echos
./start.sh # Or: npm install && npm run start
# start.sh automatically checks for Docker, Node.js, and installs dependencies
# 2. Create account at http://localhost:3000/signup
# Get your API key (authenticates to YOUR local server at localhost:4000)
# 3. Install in your project
npm install @echoshq/runtime
# 4. Initialize your workflow (2-minute setup)
npx echos init
# Choose from pre-built templates:
# Customer Support - Database + Search + Analysis
# Stripe Billing - API Integration + Analysis
# Data Analyst - SQL + Analysis
# Custom - Build your own
# 5. Run your agent
npx echos "Analyze customer sentiment from database"
# Or use in code
import { EchosRuntime } from '@echoshq/runtime';
const runtime = new EchosRuntime({
apiKey: process.env.ECHOS_API_KEY,
apiUrl: process.env.ECHOS_API_URL || 'http://localhost:4000'
workflow: './workflow.yaml' // optional
});
await runtime.run('Your task');View traces at http://localhost:3000
./start.sh # Everything on localhost - your data stays local# Option 1: Docker Compose (recommended)
docker-compose up -d
# Option 2: Deploy to your infrastructure
# Works with Railway, AWS ECS, GCP Cloud Run, Azure, Kubernetes
# Your Echos instance runs in YOUR environmentNeed help with production deployment? Join our Discord for support.
Echos fully supports AWS Bedrock for organizations that require AWS-only infrastructure.
Add these environment variables to use Bedrock:
# Set LLM provider to Bedrock
LLM_PROVIDER=bedrock
# Bedrock configuration
BEDROCK_REGION=us-east-1
BEDROCK_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
AWS_ACCESS_KEY_ID=your_access_key
AWS_SECRET_ACCESS_KEY=your_secret_key# workflow.yaml - Configure your agents to use Bedrock
config:
llmProvider: bedrock
bedrockRegion: us-east-1
bedrockModel: anthropic.claude-3-sonnet-20240229-v1:0
agents:
- name: orchestrator
maxLoops: 3
- name: db_agent
connection: ${DATABASE_URL}
policy:
guardrails:
allowedTables: [users, orders]
allowedOperations: [SELECT]
routes:
orchestrator:
canCall: [db_agent, data_agent]- Memory - How agents share context
- Embedding - Integrate into your app
- Examples - Pre-built templates
- workflow.yaml - Configuration reference
| Command | Description |
|---|---|
./start.sh |
Start Echos locally (PostgreSQL + server + web) |
./stop.sh |
Stop everything |
./restart.sh |
Restart everything |
npx echos init |
Initialize a new workflow with templates |
npx echos "task" |
Run a task with your agents |
npx echos --help |
Show CLI help |
FSL-1.1-MIT - See LICENSE