> ## Documentation Index
> Fetch the complete documentation index at: https://praison.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Privacy & Data Handling

> What happens to your prompts, code, and data when you run PraisonAI agents

PraisonAI runs on your machine and sends your prompts and data only to the LLM provider and tools you configure — the project itself never receives them.

```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph LR
    A[Your Code & Prompts] --> B[PraisonAI Agent<br/>runs locally]
    B --> C[Your Chosen LLM Provider]
    B --> D[Tools You Enable]
    B -.->|opt-in only<br/>anonymous counts| E[Telemetry]

    classDef agent fill:#8B0000,stroke:#7C90A0,color:#fff
    classDef tool fill:#189AB4,stroke:#7C90A0,color:#fff
    classDef output fill:#10B981,stroke:#7C90A0,color:#fff
    classDef optional fill:#F59E0B,stroke:#7C90A0,color:#fff

    class A,B agent
    class D tool
    class C output
    class E optional
```

## Quick Start

**1. Keep data on your machine**

Point the agent at a local model — nothing leaves your network.

```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent

agent = Agent(
    instructions="Review this proprietary code.",
    llm="ollama/llama3.2",  # runs locally
)
agent.start("Explain what this function does: ...")
```

**2. Guarantee telemetry is off**

Telemetry is already off by default. Set this to make the opt-out explicit and override any programmatic enable:

```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
export DO_NOT_TRACK=true
```

***

## Where Your Data Goes

| Data | Where it goes | Who controls it |
| - | - | - |
| Prompts, instructions, responses | The LLM provider set in `llm=` | You — read that provider's data policy |
| Files, code, tool arguments | Only the tools you give the agent | You |
| Memory, sessions, knowledge | Local storage or the database you configure | You |
| Anonymous usage counts | PostHog (EU), **only if you opt in** | You |

PraisonAI is a pass-through library: it does not proxy, store, or inspect your content on any PraisonAI-operated server. The data policy that applies to your prompts is the policy of the LLM provider you choose.

```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph TD
    Q{Can proprietary data<br/>leave your network?}
    Q -->|No| L[Use a local model<br/>Ollama, LM Studio, vLLM]
    Q -->|Yes, with a vetted vendor| V[Use a provider whose<br/>data policy you accept]

    classDef input fill:#6366F1,stroke:#7C90A0,color:#fff
    classDef output fill:#10B981,stroke:#7C90A0,color:#fff

    class Q input
    class L,V output
```

***

## Telemetry

Telemetry is **disabled by default**. It only runs if you set `PRAISONAI_TELEMETRY_ENABLED=true` (or `PRAISONAI_PERFORMANCE_ENABLED=true`) or call `enable_telemetry()`.

> [accordion] What is sent when enabled
>
> * Event counts: agent runs, task completions, tool calls, errors, feature usage
> * Tool names and execution times (no arguments), error types (no messages), success flags
> * SDK version, Python version, OS type
> * A random session ID regenerated each run
>
> GeoIP lookup is disabled.

> [accordion] What is never sent
>
> * Prompts, instructions, or LLM responses
> * Source code, files, or tool arguments and results
> * API keys or credentials
> * Personal data

> [accordion] How to force it off
>
> Any of these wins over every enable setting:
>
> ```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
> export DO_NOT_TRACK=true
> export PRAISONAI_TELEMETRY_DISABLED=true
> export PRAISONAI_DISABLE_TELEMETRY=true
> ```

See [Telemetry](https://praison.ai/docs/features/telemetry.md) for details.

***

## Best Practices

> [accordion] Review your LLM provider's data policy
>
> Your prompts go straight to the provider. Check its retention and training-use terms, or use an enterprise/zero-retention plan.

> [accordion] Use a local model for sensitive code
>
> Set `llm=` to an Ollama or other self-hosted model so content never leaves your infrastructure.

> [accordion] Only give agents the tools they need
>
> Web search, MCP servers, and API tools send their arguments to third parties. Leave them off for confidential work.

> [accordion] Set DO\_NOT\_TRACK in CI and shared environments
>
> It guarantees no telemetry even if a library or script calls `enable_telemetry()`.

***

## Related

> [card] [Telemetry](https://praison.ai/docs/features/telemetry.md)
>
> What anonymous metrics are collected and how

> [card] [Context Redaction](https://praison.ai/docs/features/context-security-redaction.md)
>
> Strip secrets from context before it reaches the LLM


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.