Streetrace

Open runtime and DSL for structured multi-agent systems

Streetrace is an open-source runtime that lets you define and execute structured multi-agent systems anywhere Python runs — locally, in Docker, or in CI/CD workflows. It is designed for advanced agent workflows where agents behave as explicit systems, with guardrails and execution constraints.

Why Streetrace Exists

Traditional AI agent tooling treats agents as informal prompt glue. Streetrace treats agents as systems. It enables:

• Explicit agent definitions via a simple DSL and Python abstractions
• Multi-agent orchestration with controlled interactions
• Execution guardrails via constraint handlers similar to structured safety tooling
• Flexible execution environments: local, cloud, CI/CD (e.g., GitHub Actions)

This makes Streetrace suitable for developers and platform engineers building structured agent systems, automation workflows, or CI-integrated agent tooling.

Quick Start

Install StreetRace, create a custom AI agent, and run it on a real task — all in 5 minutes.

Get Started →

What It Does Today

📌 Core Features

• Python-based runtime that works locally or in CI/CD
• DSL agent definitions for structured workflows
• Multi-agent orchestration with explicit execution semantics
• Constraint handling for safe executions (work in progress but usable)
• Tool integrations for common tasks (filesystem, CLI, search, etc.)

🚫 Not Included (Yet)

This project does not yet provide:

• Built-in evaluation frameworks
• Versioned agent lifecycle management
• Centralized fleet or policy plane

(These are future roadmap items.)

Example Agent Definition

model main = anthropic/claude-sonnet
model gpt = openai/gpt-5.2

tool fs = builtin streetrace.fs
tool github = mcp "https://api.githubcopilot.com/mcp/" with auth bearer "${GITHUB_PAT}"

retry default = 3 times, exponential backoff
timeout default = 2 minutes

on input do
    mask pii
    block if jailbreak
end

on output do
    mask pii
end

prompt analyze_code using model gpt: """You are a historical context analysis expert. Analyze the historical context for the provided codebase."""

prompt main_instruction: """You are a code analysis assistant. Help users analyze their codebase for quality issues. Available commands: Analyze a file or directory, Get recommendations for improvement, Explain specific issues."""

agent code_analyzer:
    tools fs, github
    instruction analyze_code
    description "Analyzes code quality"

agent:
    tools fs
    instruction main_instruction
    use code_analyzer

This file defines a structured agent with specific tool bindings.

Usage Scenarios

Server Deployment

Use Streetrace in server environments with session management:

# Create .env file with required environment variables
cat > .env << 'EOF'
${dockerEnvVars}${dockerEnvVars ? '\n' : ''}
# Configure Streetrace
STREETRACE_API_KEY=<optional-streetrace.ai-key>
STREETRACE_AGENT_ID=<agent-path-url-or-streetrace_id>
# Optional user prompt
STREETRACE_PROMPT=<your-value-here>
EOF

docker run --env-file .env streetrace/streetrace:latest

CI/CD

Integrate agents into automated workflows such as GitHub Actions.

Run with Streetrace Cloud defined agent:

name: Streetrace Agent
on: [push]
jobs:
  run-agent:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
      - id: streetrace
        uses: streetrace-ai/github-action@main
        with:
          claims: |
            agent_name: "code_analyzer"
          prompt: YOUR PROMPT TO TRIGGER ANALYSIS.
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          STREETRACE_API_KEY: ${{ secrets.STREETRACE_API_KEY }}

Or run your own agent defined in your repo:

name: Streetrace Agent
on: [push]
jobs:
  run-agent:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
      - uses: actions/setup-python@v6
        with:
          python-version: '3.11'
      - name: streetrace
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          STREETRACE_API_KEY: ${{ secrets.STREETRACE_API_KEY }}
        run: |
          pip install streetrace
          streetrace --agent=./agents/code_analyzer.sr --prompt "Check new PR changes"

Local Dev

Run agents interactively as part of your development workflow:

cd myproject
streetrace --model=gpt-4o

Tools Included

All tools are sandboxed within the working directory.

Docs and Next Steps

See the docs/ folder for:

• Backend configuration
• Model provider setup
• Tool usage
• Redis caching setup

Link these pages so users can quickly go deeper.

Contributing

We welcome contributions. Before submitting a PR:

1. Write or update tests
2. Include documentation for new features
3. Follow existing styling and patterns

Check the CONTRIBUTING.md file for details.

License

Streetrace is released under the MIT License.