# Staso AI

> Observability, enforcement, and evaluation for AI agents in production.

Staso AI is the infrastructure layer for AI agents after deployment. We give engineering teams full visibility into what their agents do, why they fail, and how to fix them — without rewriting agent code.

Website: https://staso.ai
Documentation: https://staso.ai/docs
Changelog: https://staso.ai/changelog

## What Staso Does

### Observability (Live)
Full execution traces for every AI agent interaction:
- **Traces**: Every agent run captured as a trace with spans for each LLM call, tool call, and subagent invocation
- **Monitoring Dashboard**: Real-time metrics — total traces, error rate, P95 latency, average latency, total tokens, total cost
- **Charts**: Trace volume, error rate, latency P50/P95, token usage, and cost by model over time
- **Conversations**: Multi-turn agent interactions grouped by session, with trace count, span count, duration, token usage, and cost
- **Trace Detail**: Span tree with waterfall timeline showing execution flow, input/output at every node, and metadata
- **Filtering**: Filter by environment, model, agent, root span, status, and time range
- **Cost Tracking**: Per-trace and aggregate cost tracking across models

### Enforcement Firewall (Coming Soon)
Runtime validation layer for every tool call before execution:
- Business rules defined as code
- Supports deterministic rules and LLM-based reasoning
- Each rule runs in audit-only or enforce mode
- Sync and async validation modes

### Input/Output Evaluation (Coming Soon)
- Input scanning: prompt injection, jailbreaks, PII leakage, credential exposure
- Output validation: hallucinations, reasoning errors, context loss, false completion claims
- Evidence-backed evaluation at every step

### Self-Healing & RCA (Coming Soon)
- Code-aware root cause analysis classifying failures by type
- Automated fix PR generation
- Agent checkpointing with before/after metric comparison

## Integration

Staso integrates with existing agent code via a Python SDK. No agent rewrites required.

### Python SDK (Decorator-based)
```python
from staso import Staso

staso = Staso(api_key="ak_...")

@staso.trace
def run_agent(query):
    return agent.run(query)

@staso.tool_call
def execute_action(tool, params):
    return tool.run(params)
```

### OpenAI Integration
```python
from openai import OpenAI
from staso.integrations import wrap_openai

client = OpenAI()
client = wrap_openai(client)  # One line. Full tracing.
```

### Anthropic Integration
```python
from anthropic import Anthropic
from staso.integrations import wrap_anthropic

client = Anthropic()
client = wrap_anthropic(client)  # One line. Full tracing.
```

### Claude Code Integration
```bash
pip install staso
staso setup-claude-code \
  --api-key ak_... \
  --workspace my-workspace \
  --environment production \
  --agent-id my-agent
```

Automatically captures every turn, LLM call, and tool call from Claude Code agents. No code changes needed.

### Supported Frameworks
- **OpenAI SDK** — Live (wrap_openai)
- **Anthropic SDK** — Live (wrap_anthropic)
- **Claude Code** — Live (CLI setup)
- **Google SDK** — Coming soon
- **LangChain** — Coming soon

### Installation
```bash
pip install staso
```

### Configuration
Environment variables:
- `STASO_API_KEY` — API key for authentication
- `STASO_BASE_URL` — Custom API endpoint (default: https://api.staso.ai)
- `STASO_WORKSPACE_SLUG` — Workspace identifier
- `STASO_ENVIRONMENT` — Environment tag (e.g., production, staging, development)
- `STASO_AGENT_ID` — Agent identifier shown on dashboard
- `STASO_DEBUG` — Enable debug logging

## Dashboard Features

### Traces View
- Searchable table of all agent executions
- Columns: Status, Name, Input, Output, Model, Agent, Latency, Tokens, Cost, Time
- Filter by environment, model, agent, status
- Click any trace to see full span tree and waterfall

### Monitoring Dashboard
- 6 key metrics: Total Traces, Error Rate, P95 Latency, Avg Latency, Total Tokens, Total Cost
- Trend indicators showing percentage change
- Interactive charts: Trace Volume, Error Rate, Latency P50/P95, Token Usage, Cost by Model
- Time range and environment filters

### Conversations View
- Group traces by session/conversation
- See multi-turn agent interactions
- Track conversation-level metrics: trace count, span count, duration, tokens, cost

### Trace Detail
- Span tree showing execution hierarchy (Agent → LLM → Tool)
- Waterfall timeline for latency analysis
- Input/Output/Metadata tabs for every span
- Cost and token breakdown per span

## Use Cases

Staso is built for engineering teams who:
- Run AI agents in production and need visibility into agent behavior
- Want to debug agent failures with execution traces and span trees
- Need cost tracking and optimization for LLM usage across models
- Are building multi-step agents with tool calls and need to trace every step
- Want to enforce guardrails on agent tool calls before execution
- Need to detect prompt injection, hallucinations, and PII leakage

## Pricing

### Free (Personal)
- 10,000 traces/month
- 7 days data retention
- Up to 5 team members
- Up to 5 API keys
- 3 workspaces

### Team (Coming Soon)
- 100,000+ traces/month
- 30 days data retention
- Unlimited team members
- Unlimited API keys
- Unlimited workspaces
- Custom roles
- Priority support

### Enterprise
- Custom trace volume
- Custom retention
- Dedicated support
- Contact: support@staso.ai

## Technical Details

- **API Base URL**: https://api.staso.ai
- **Dashboard**: https://staso.ai/dashboard
- **Authentication**: OAuth (Google, GitHub) for dashboard; API keys for SDK
- **Data Format**: OpenTelemetry-compatible trace format
- **Supported Languages**: Python (SDK available via pip)

## Company

- Founded by Mayank Sheoran and Kunal Singal
- YC-backed startup
- Website: https://staso.ai
- Support: support@staso.ai
- Documentation: https://staso.ai/docs
- Changelog: https://staso.ai/changelog
- About: https://staso.ai/about

## Documentation Pages

- Getting Started: https://staso.ai/docs
- Quickstart Guide: https://staso.ai/docs/getting-started/quickstart
- Tracing Guide: https://staso.ai/docs/guides/tracing
- Conversations Guide: https://staso.ai/docs/guides/conversations
- OpenAI Integration: https://staso.ai/docs/integrations/openai
- Anthropic Integration: https://staso.ai/docs/integrations/anthropic
- Claude Code Integration: https://staso.ai/docs/integrations/claude-code
- Configuration Options: https://staso.ai/docs/configuration/options
- Troubleshooting: https://staso.ai/docs/troubleshooting