# AgentLeak

> Open-source privacy testing for single-agent and multi-agent systems. AgentLeak analyzes complete execution traces across tool calls, shared memory, inter-agent messages, logs, generated files, and final output.

An agent can return a clean answer while private data leaves through a channel nobody reads. Measured across the 283 scenarios bundled with the tool, every leak behaves that way: it travels on an internal channel and never reaches the final output. Scoring is deterministic — exact ground-truth matching, no judge model — so the same trace always yields the same number.

AgentLeak can be used by developers through its CLI and Python SDK, or directly by autonomous agents through a project-scoped HTTP API. Prefer synthetic or canary data for tests. The hosted service is available at https://www.agentleak.org.

## Start here

- [https://www.agentleak.org/docs](https://www.agentleak.org/docs): Documentation hub and product model
- [https://www.agentleak.org/benchmark](https://www.agentleak.org/benchmark): Measured internal-channel leakage and detection-tier coverage across the 283 bundled scenarios, with the command to reproduce it
- [https://www.agentleak.org/compare](https://www.agentleak.org/compare): How AgentLeak differs from promptfoo, Garak and PyRIT — they test the model, this tests the system
- [https://www.agentleak.org/compliance/eu-ai-act](https://www.agentleak.org/compliance/eu-ai-act): EU AI Act Annex III evidence, mapped to Arts. 9, 11, 12, 15 and 26
- [https://www.agentleak.org/docs/developers](https://www.agentleak.org/docs/developers): CLI, SDK, integrations, trace schema, and CI guide
- [configuration reference](https://github.com/yagobski/agentleak/blob/main/docs/configuration.md): complete agentleak.yaml reference
- [CLI reference](https://github.com/yagobski/agentleak/blob/main/docs/cli.md): commands, options, exit codes and artifacts
- [red-team quickstart](https://github.com/yagobski/agentleak/blob/main/docs/redteam-quickstart.md): campaign matrix, modes, metrics and safety
- [https://www.agentleak.org/docs/red-team](https://www.agentleak.org/docs/red-team): Hosted red-team getting started guide
- [https://www.agentleak.org/docs/red-team/architecture](https://www.agentleak.org/docs/red-team/architecture): Generation, target, trace, evaluation and evidence architecture
- [https://www.agentleak.org/docs/red-team/llm-vulnerability-types](https://www.agentleak.org/docs/red-team/llm-vulnerability-types): F1-F6 LLM and agent vulnerability taxonomy
- [https://www.agentleak.org/docs/red-team/plugins](https://www.agentleak.org/docs/red-team/plugins): Live native and Promptfoo-compatible plugin catalog
- [https://www.agentleak.org/docs/red-team/strategies](https://www.agentleak.org/docs/red-team/strategies): Delivery strategies, profiles and attack matrices
- [https://www.agentleak.org/docs/privacy-compliance](https://www.agentleak.org/docs/privacy-compliance): Privacy assurance levels, finding-to-control evidence matrix, governance assertions and CI workflow
- [reporting guide](https://github.com/yagobski/agentleak/blob/main/docs/reporting.md): report contract, formats and redaction
- [https://www.agentleak.org/docs/agents](https://www.agentleak.org/docs/agents): Autonomous onboarding, safety rules, self-test loop, and failure handling
- [https://www.agentleak.org/docs/api](https://www.agentleak.org/docs/api): Integrated API reference with auth, endpoint flows, examples, and schema notes
- [https://www.agentleak.org/agents.md](https://www.agentleak.org/agents.md): Normative instructions for autonomous agents

## Machine interfaces

- [https://www.agentleak.org/openapi.json](https://www.agentleak.org/openapi.json): OpenAPI 3 schema
- [https://www.agentleak.org/api/schemas](https://www.agentleak.org/api/schemas): JSON Schema catalog for config, traces, reports, policies, code scans, red-team requests, and agent cards
- [https://www.agentleak.org/api/docs](https://www.agentleak.org/api/docs): Swagger UI fallback for raw schema exploration
- [https://www.agentleak.org/api/meta](https://www.agentleak.org/api/meta): Runtime capabilities, channels, limits, and endpoint map
- [https://www.agentleak.org/.well-known/agent-card.json](https://www.agentleak.org/.well-known/agent-card.json): Agent capability manifest with a documented custom HTTP+JSON binding
- [https://www.agentleak.org/llms-full.txt](https://www.agentleak.org/llms-full.txt): Complete agent-oriented usage guide in one document

## Optional

- [https://github.com/yagobski/agentleak](https://github.com/yagobski/agentleak): Source code and self-hosting