Streetrace
Open Source

AI Agent Infrastructure.Governance
Built-In.

Where any knowledge worker builds agents that are automatically auditable, policy-compliant, and governed by design.

U
Product Manager
"Check website stats daily and summarize changes. Make sure to mask any PII before analyzing."
Generating Agent System...
system.streetrace
AUTO-GENERATED
agent StatsReviewer:
  tools fetch_analytics

on input do
  mask pii
end

flow DailyReview:
  run StatsReviewer
Agent Deployed
Active & Governed
Open-Source Runtime
100+ LLMs via LiteLLM
Runs on your infra
Four layers of guardrails
Full audit trail
Centralized Policies, Delegated Ownership

You own the policies.
Your teams own the AI.

Develop strict rules, permissions, and guardrails centrally, so anyone can deploy agents reliably.

A product manager says “Check website stats daily and summarize changes,” and Streetrace automatically generates and deploys the complete fully-compliant agent system. Vertical teams move fast while horizontal teams retain visibility and control.

Centralized Policies

Product Teams

Compliant AI Agent

Ready & Governed

Deep Data Flow Control

Four Layers of Guardrails.
Zero Gaps.

Most platforms filter inputs and outputs. Streetrace enforces control at every layer of agent execution.

Compile-Time

Zero Latency

Structural constraints enforced before agents run. Blocked code paths structurally do not exist.

Run-Time Deterministic

Pattern-based

Entity recognition, classification, or pattern matching. Mask data before the model.

Run-Time AI Validation

High Accuracy

A second model validates critical decisions against your policies before execution continues.

Conversation Steering

Goal-anchored

Detects drift from user manipulation or model tendency, steers back to the authorized task.

Agents as Systems, Not Prompt Glue

Your entire agent architecture. One file.

Streetrace DSL defines agents, guardrails, data flows, schemas, and multi-agent workflows in a single declarative file. Your AI system is one artifact—reviewable, deployable, auditable.

risk_policy.streetrace
model main = anthropic/claude-opus-4-6

// Enforce response format via schema
schema RiskOutput:
  risk_score: int
  findings: list[string]

prompt auditor_prompt expecting RiskOutput using model "main": """
  Analyze the provided context for compliance.
"""

agent Auditor:
  instruction auditor_prompt

Version it in Git

Treat AI like code. Every change is a commit. Every rollout is a PR.

Infrastructure Isolation

Cleanly separate agent logic from application code. Evaluate and manage your AI as a self-contained system.

Model Agnostic by Design

Connect 100+ LLMs via LiteLLM. Swap providers without touching agent definitions; your guardrails automatically follow.

EU AI Act Compliance

Executable Compliance for the
August 2026 Mandate

While others draft policy documents, Streetrace users deploy compliant architecture. Explore how our DSL and Open-Source Engine satisfy the high-risk requirements of the EU AI Act.

01Engine Trait

Article 5: Prohibited AI Practices

The EU AI Act prohibits AI systems that deploy subliminal techniques, exploit vulnerabilities, or are used for social scoring.

Streetrace enforces Compile-Time (Zero Latency) and Run-Time Deterministic guardrails that block specific use cases or intents before execution. Centralized policies can be formally codified into the agent’s architecture to ensure prohibited practices are systematically blocked at the data flow level, without relying on scattered prompt instructions.

The Streetrace Difference

From Shadow AI to Governed Infrastructure

Today's Reality

Agent code tangled in application repos, prompts, and APIs.

Tracing compliance requires weeks of parsing raw application logs.

PII leaks and unmonitored tool calls bypass compliance teams.

Stuck with one model's ecosystem and safety wrappers.

Streetrace Architecture

Entire agent system defined in one reviewable, deployable DSL file.

Full audit trail generated automatically on every single agent run.

Policy enforcement on every model input, output, tool call, and RAG query.

100+ LLMs via LiteLLM—swap anytime, guardrails follow instantly.

Predictable Operations

Open-Source Runtime.
Absolute Data Sovereignty.

Streetrace agent runner executes your DSL systems anywhere Python runs. Deploy as Docker containers, GitHub Actions, or locally.

100% Data Control

Telemetry is OTEL-native. Ship to Streetrace Cloud for enhanced analytics and reasoning transparency, or redirect your audit trails and analytics to Datadog, Splunk, Langfuse, or Arize via standard OTEL_EXPORTER_OTLP_ENDPOINT config. Keep it entirely air-gapped or fully connected.

Audit-Ready by Design

Every decision made, tool invoked, and guardrail evaluated is logged and full structured logs immediately available.

Customer VPC / Air-gapped

Streetrace Agent Runner

Docker Container

OTEL Telemetry
Datadog
Splunk
Langfuse

Ready to scale AI
while managing the risk?

See how FinTech leaders use Streetrace to deploy autonomous, fully-governed multi-agent systems before the August 2026 deadline.