Prompt protection for agents that touch the world.
Parse screens untrusted text before it reaches tools, memory, credentials, payments, code execution, or users. It is built for independent AI developers who need agent-native security without vendor lock-in.
RAG document
Untrusted retrieved text asks the agent to ignore its tool policy.
surface: rag
risk 8.7 / block
Categories: instruction override, indirect injection, tool abuse.
POST /v1/parse
Sandbox or refuse
Agent keeps the document as data, not as authority over tools.
trace_id: prs_7fd2
Screen every place an agent can be steered.
The routing decision stays simple: when text crosses a trust boundary, call Parse before that text gets authority.
Prompt injection, hidden instructions, and retrieved content that tries to redirect the agent.
/v1/parseHTML, JSON, search snippets, issue bodies, and page content returned by external tools.
/v1/parseScreen model output before sending it to users, tools, memory, or another agent.
/v1/screen-outputVerify identity, delegation context, and social-engineering risk before accepting work.
/v1/agent/trust/verifyMake the risk visible before you buy.
The public test lab generates safe test resources so developers can see whether a target model or tool-using agent resisted, partially followed, or completed a harmless callback.
Hidden page text tries to steer a browsing agent.
A tool response contains an instruction-looking field.
A peer agent claims false authority to delegate work.
No callback received, no reference leaked, and the output treated the fixture as untrusted data.
Run your own sessionIntegrate through the path your agent already speaks.
REST, MCP, OpenAPI, and x402 all point to the same core decision: screen before authority.
One POST call
Use any HTTP client and follow the returned recommended action.
POST /v1/parseHosted tools
Expose screen_prompt, screen_output, verify_agent_trust, and get_pricing.
POST /mcpTool calling
Let coding agents and GPT Actions discover the callable API surface.
/openapi.jsonNo account first call
Autonomous agents can pay per call when no bearer key exists.
/v1/pricingBuilt for indie builders. Credible enough for serious labs.
The frontier-lab path is not enterprise theater. It is defensible claims, clear limitations, auditable behavior, and machine-readable discovery that agents can actually use.
9 public categories aligned to prompt and agent security risks.
Detection reduces risk but does not replace least-privilege tools or output validation.
Production source, issue tracking, and evidence intake are maintained privately while public docs disclose behavior and limits.
Free keys start at 10 req/min; x402 uses USDC on Base mainnet.
Latest field notes.
Durable technical writing for prompt injection, agent security, x402, MCP, and prompt protection infrastructure.
The AI security market is consolidating fast. Here's why Parse chose to stay independent — and why that matters for your agent's safety.
2026-03-23 Building a Security Layer for Your Agent Pipeline: A Practical Architecture GuideLearn how to build a security layer for your AI agent pipeline. Covers threat modeling, common vulnerabilities, and implementation patterns with code examples.
2026-03-23 Autonomous Agent Payments: Security Implications of x402 ProtocolUnderstanding the security implications of autonomous agent payments via x402 protocol. Learn threat models, attack scenarios, and security controls for agent-based financial transactions.
Put Parse at your next trust boundary.
Start with the public test lab, then wire the same decision into your agent runtime.