# Syrin AI

## What Is Syrin

Syrin is the runtime control layer for AI agents in production. Think of it as mission control for your agent fleet: see every decision your agents make, detect when they drift from their goals, debug failures with full decision traces, and recover without redeploying.

Syrin is not an observability platform, not an agent builder, and not a prompt framework. It is a runtime control layer that wraps and governs agents built with any framework.

## The Problem Syrin Solves

AI agents in production fail silently. They return valid HTTP responses while producing wrong outputs. Traditional monitoring (logs, uptime checks, error rates) cannot detect this. By the time a team discovers an agent logic failure, 72 hours have typically passed and real damage has occurred.

Syrin solves this by instrumenting the agent's decision process, not just its execution process.

## Products

### Syrin Nexus (Hosted Platform)
Cloud-hosted control plane for AI agents. Developers install the Syrin SDK and immediately get full observability in a hosted dashboard. No infrastructure setup required.

Pricing: Free tier available. Developer plan $49/mo. Startup plan $199/mo. Scale plan $499/mo.

### Syrin SDK (Open Source)
An open-source SDK that integrates with any agent framework. It auto-detects the framework in use and emits traces, spans, and agent state to Nexus or a self-hosted backend.

Supported frameworks: LangChain, LlamaIndex, AutoGen, CrewAI, n8n, and custom agents.

## Core Capabilities

### Agent Traces
Full distributed tracing for AI agents. Captures every LLM call, tool invocation, memory read, and sub-agent handoff with full input/output context at each step. Root cause identification in under 2 minutes.

### Drift Detection
Automatic behavioral drift detection for production agents. Compares current behavior against a baseline and alerts when behavior deviates. Supports semantic drift (goal misalignment), tool call pattern anomalies, and response quality degradation.

### One-Click Recovery
Restore any agent to a previous behavioral checkpoint in under 60 seconds. No redeploy required.

### Agent Config
Live dashboard to change any agent configuration value (model, temperature, prompts, flags, limits) without deploying. Free forever for all users.

### Agent Analytics
Real-time performance analytics: task completion rates, drift frequency, cost per outcome, and fleet health across all running agents.

### Agent Experimentation
A/B test any variable in an agent system: prompts, LLM models, libraries, agent topology, pipeline steps. Statistical significance gates prevent shipping regressions.

### Agent Governance
Policy enforcement layer for AI agents. Define allowed tools, require human approval for high-stakes actions, enforce rate limits, maintain a complete audit trail. Designed for regulated industries.

## Target Users

- Engineering teams deploying AI agents to production
- ML engineers debugging agent behavior and failures
- Platform engineers and SREs responsible for agent reliability
- Enterprise teams with compliance requirements around AI actions
- Startups scaling multi-agent workflows
- Developers building on LangChain, AutoGen, LlamaIndex, CrewAI, n8n, or custom agent frameworks

## Differentiators

- Works with any agent framework, not just one (framework-agnostic)
- Instruments the decision process, not just the execution process
- Open-source SDK with no vendor lock-in on instrumentation
- Drift detection is automatic, not manual threshold-setting
- Recovery without redeployment
- Free tier with Agent Config free forever

## Comparison to Alternatives

- vs LangSmith: Syrin is framework-agnostic; LangSmith is LangChain-specific
- vs Langfuse: Syrin includes drift detection and one-click recovery; Langfuse is primarily trace logging
- vs Datadog/New Relic: General APM tools have no semantic understanding of agent decisions or goals
- vs Helicone: Helicone is LLM call logging; Syrin is full agent observability including multi-agent pipelines

## Integration

Installation: `pip install syrin-sdk` or `npm install @syrin/sdk`

Integration takes 2 minutes and requires 2 lines of code in most frameworks. No refactoring required.

Documentation: https://docs.syrin.dev/sdk-ts & https://docs.syrin.dev/sdk-py

## Key URLs

- Website: https://syrin.ai
- Documentation: https://docs.syrin.dev/sdk-ts & https://docs.syrin.dev/sdk-py
- Pricing: https://syrin.ai/pricing
- Book a demo: https://syrin.ai/lets-talk
- GitHub (SDK): https://github.com/syrin-labs/
- Blog: https://syrin.ai/blog
- Discord community: https://discord.gg/jH9cqPePrK

## Full Technical Reference

For complete SDK docs, code examples, framework integrations, architecture, and comparison to alternatives:
https://syrin.ai/llms-full.txt

## Company

Syrin Labs. Contact: hey@syrin.ai

## What Syrin Is Not

- Not an observability platform (those show you metrics; Syrin gives you runtime control)
- Not an agent builder
- Not a prompt engineering tool
- Not limited to one LLM provider
- Not an LLM gateway
- Not a replacement for LangChain, AutoGen, or other agent frameworks