Privacy Policy
Last updated: August 11, 2026
Overview
Parse ("we," "our," or "us") respects your privacy and is committed to protecting your personal data. This privacy policy explains how we collect, use, disclose, and safeguard your information when you use our prompt security API service.
Information We Collect
1. API Request Data
We process prompts and content you submit to the Parse API for security analysis. This includes:
- Prompts: Text content submitted for injection screening and safety analysis
- Metadata: Optional fields like agent_id, session_id, and source identifiers
- Execution Results: Sandbox outputs when execute: true is enabled
2. API Keys and Authentication
We collect and store:
- API Keys: Self-generated keys via POST /v1/keys/generate (hashed for storage)
- Key Metadata: Name, creation date, expiration date, scopes, and usage statistics
- IP Addresses: For rate limiting and abuse prevention. Rate-limit counters key off a SHA-256 of the address and expire with the window; the audit log records the address itself, under the retention below.
3. Payment Data (x402)
When using x402 USDC payments:
- We do not store payment credentials or private keys
- Payment verification occurs on-chain via the Base L2 blockchain
- We record transaction hashes for billing and audit purposes
How We Use Your Data
1. Service Delivery
We use your data to:
- Analyze prompts for security threats (prompt injection, jailbreaks, adversarial patterns)
- Return risk scores, flags, and safety assessments
- Execute prompts in isolated sandbox environments when requested
- Enforce rate limits and prevent abuse
2.Service Improvement
We may use anonymized, aggregated data to:
- Improve detection accuracy and reduce false positives
- Identify new attack patterns and threat vectors
- Optimize API performance and latency
- Generate usage statistics and analytics
What We Store, Per Endpoint
Storage does not vary by plan. Free, Pro, Team, and Compliance keys are handled identically — the tier changes rate limits, cost caps, and which fields come back in the response, not what Parse writes down.
| Endpoint | Is your prompt text stored? | What Parse records |
|---|---|---|
POST /v1/parse, POST /v1/screen-output, POST /v1/agent/trust/verify |
No. The screening event table has no column for prompt or output text, and none for a hash of it. | Risk score, verdict, categories, screening mode, latency, blocked flag, enforcement mode, request ID, matched rule IDs, the metadata labels you send (source_kind, trust_level, intended_action), API key ID, timestamp. |
POST /v1/evaluate |
Yes, while the run is in flight. When the evaluation ends, successfully or not, Parse overwrites its copy with the first 100 characters of your prompt plus a SHA-256 of the whole prompt. Those first 100 characters stay readable. | Evaluation results, model name, token counts, cost, and the redacted prompt described to the left. |
| Audit log (written by every screened call) | No. The prompt's length is recorded as a number; the text is not. | Action, API key ID, risk score, verdict, prompt length, categories, rule IDs, request ID, and the caller's IP address. |
| Compliance receipts | No. | Verdict, risk score, matched rule IDs, agent ID, policy version, and the receipt hash chain. |
| API keys | Not applicable | A bcrypt hash plus a lookup prefix. The full key is never written down; it is shown once at generation. |
Data Retention
| Record | Stated retention | How it is enforced today |
|---|---|---|
| Screening events | 90 days | Automatic. A daily job deletes records past the window. |
| Numbers-only daily rollups | Kept indefinitely | Counts and latency percentiles only, no prompt-derived content. Written by the daily job before the raw events they summarize are deleted, so detection-improvement history survives the purge. |
| Audit events, including the caller IP | 90 days | Automatic, as above. |
| Compliance receipts | 1 year, fixed so the hash chain stays verifiable | Automatic, as above. |
Redacted /v1/evaluate records | The 500 most recent, then dropped | Automatic. They are held in the server's memory, so a restart clears them. |
| Rate-limit counters in Redis | The length of the rate-limit window | Automatic, via Redis key expiry. |
| API keys | Until revoked, or the expiry set when the key was made (90 days by default for self-service keys) | Automatic on expiry. |
Enforcement is automated. A daily purge job deletes screening events, audit events, and compliance receipts past their stated windows. To request early removal, email [email protected] — we complete deletion requests within 30 days.
How screening metadata is used, and how to opt out. We retain structured, numbers-only metadata (verdict counts, category distributions, rule-hit rates, latency percentiles — never prompt or output text) to improve detection and prevent abuse. If you prefer your traffic be excluded from detection-improvement aggregates, email [email protected] with your key name and we will exclude it within 7 days. Exclusion changes nothing about how your requests are served, rate-limited, or protected — it only removes your traffic from the aggregate improvement statistics.
Where Your Prompt Text Goes
Screening runs on Parse's own infrastructure. Prompt text leaves it in three cases, all of them listed here.
| Recipient | When it receives your prompt | How to prevent it |
|---|---|---|
| OpenRouter — routes the semantic analysis layer to a model provider | On POST /v1/parse and POST /v1/screen-output, the text is sent for scoring. Parse skips the call when you pass mode: "pattern-only", when a pattern already matched at severity 9 or above and settles the verdict, or when the deployment has no OpenRouter key configured. |
Pass "mode": "pattern-only" on the request. |
| OpenRouter — runs the prompt against a model | Only when you pass execute: true, which asks Parse to run the prompt on purpose and screen what comes back. |
Omit execute. It is off by default. mode: "pattern-only" does not turn it off. |
| The execution sandbox — an isolated runner, configured per deployment | Only on the same execute: true path, which sends the prompt and any test_input. |
Omit execute. |
| Stripe | Never. Stripe sees subscription and payment metadata. Card details go to Stripe directly and Parse never holds them. | — |
What OpenRouter and the model providers behind it do with text they receive is governed by their policies, not ours. If that matters to you, use pattern-only mode and the text never reaches them.
POST /v1/parse
{
"prompt": "...",
"mode": "pattern-only"
}Pattern-only screening stays entirely on Parse infrastructure, and it is a real
trade: pattern matching alone under-reports paraphrased and indirect attacks that
the semantic layer catches. Parse does not hide the trade — every response reports
which layers ran, and a pattern-only response carries
layers.llm: "skipped_pattern_only".
Data Security
We implement industry-standard security measures:
- Encryption: All data in transit uses TLS 1.3
- API Keys: Stored as SHA-256 hashes (plaintext keys shown once at generation)
- Sandbox Isolation: Execution environments are containerized with no network access
- Access Controls: Strict authentication and authorization for all internal systems
- Monitoring: 24/7 intrusion detection and security auditing
Data Sharing and Disclosure
We do not sell your data. We may share data only in the following circumstances:
1. Service Providers
- Infrastructure: Application hosting, Postgres database, and Redis cache
- Payment Processing: Stripe (subscriptions) and x402 protocol facilitators (verify on-chain USDC transfers)
- AI Models: OpenRouter, which routes the semantic analysis layer to providers such as DeepSeek, OpenAI, and Anthropic
OpenRouter is the only one of these that receives your prompt text, under the conditions set out in Where Your Prompt Text Goes above. What it and the model providers behind it retain is governed by their policies, not ours. Pass mode: "pattern-only" and the text never reaches them.
2. Legal Requirements
We may disclose data if required to:
- Comply with legal obligations (court orders, subpoenas, warrants)
- Protect our rights, property, or safety
- Prevent fraud, abuse, or security threats
- Enforce our Terms of Service
Your Rights and Choices
1. API Key Management
- Generate: Create new API keys via POST /v1/keys/generate
- Revoke: Delete keys by contacting support (keys expire after 30 days by default)
- Configure: Adjust screening policy via GET/PUT/DELETE /v1/policy
2. Data Access and Deletion
- Usage Records: Request a copy of your API usage history by contacting support
- Deletion: Request deletion of your account and associated data (retained for legal/audit requirements)
- Export: Export your screening policies and configuration data at any time
3. Opt-Out
- Analytics: We do not use third-party analytics cookies or trackers
- Marketing: We do not send marketing emails or use your data for advertising
GDPR and UK Data Protection
Parse for Agents processes personal data on behalf of its customers as a data processor under Article 28 of the GDPR. Our Data Processing Agreement is available at /dpa.
Lawful basis
We process personal data under the lawful bases of contract (providing the screening service) and legitimate interests (security, fraud prevention, network integrity).
International data transfers
Personal data may be transferred from the EEA/UK to the United States under the Standard Contractual Clauses (SCCs). See our DPA for the full transfer mechanism and a Transfer Impact Assessment summary.
Data residency
Processing currently occurs in the United States. An EU/UK region is on our roadmap. Parse cannot offer EU data residency today. mode: "pattern-only" prevents onward transfer of prompt text to the semantic-analysis sub-processor (OpenRouter, US), but prompt text is still transferred to Parse for processing in the United States.
Sub-processors
See the sub-processor list on our Trust page or the full DPA sub-processor table with GDPR adequacy status.
Your rights
You have the right to access, rectify, erase, restrict processing of, and port your personal data. To exercise these rights, contact [email protected].
Data Protection Officer
For data protection inquiries, contact [email protected].
Children's Privacy
Parse is not intended for children under 13. We do not knowingly collect personal information from children under 13. If we become aware of such collection, we will delete it promptly.
International Data Transfers
Parse may store and process data in the United States and other countries where our service providers operate. By using our service, you consent to this transfer, processing, and storage of your data.
Changes to This Privacy Policy
We may update this privacy policy from time to time. We will notify you of material changes by:
- Posting the new policy on our website
- Updating the "Last updated" date at the top of this policy
- Sending a notification to your registered email (if provided)
Your continued use of Parse after changes constitutes acceptance of the new policy.
Contact Us
For questions about this privacy policy, your data, or your rights, contact us:
- Email: [email protected]
- Website: https://www.parsethis.ai