Secrets and sensitive data don’t just live in text.
Transmute is an LLM middleware transformation engine. Detect and de-identify secrets and sensitive data across supported text, documents, images, and attachments. Transform the outbound content, preserve useful context, and optionally require human review before sending to your AI provider.
Keep captured data in your own infrastructure with supported cloud and storage connectors. You choose the destinations and retention policies. See what leaves your deployment.
The useful stuff stays. The sensitive stuff gets handled.
A face in an image. An identity in a document. See the difference before anything leaves for AI.
Keep the meaning. Replace the identifiers.
NER + coreference + typed tokens
Before · sample.pdfAfter · transformed preview
PURFECT LABS DEMO DOCUMENT
Account review
Internal summary / 08 October 2026
Prepared for
Morgan Ellis
Email
morgan@example.com
Account
4821 0937
Review findings
Morgan Ellis reported a duplicate charge. The transaction appeared twice on the October statement.
Riley Chen reviewed the statement. The reviewer asked Morgan to confirm the date.
The account holder confirmed October 3. Riley recommended a reversal.
Recommended action
Compare the transaction timestamps and confirm whether a reversal has already been issued.
PURFECT LABS DEMO DOCUMENT
Account review
Internal summary / 08 October 2026
Prepared for
{{REDACTED_PERSON_9b64a}}
Email
{{REDACTED_EMAIL_9b64a}}
Account
{{REDACTED_ACCOUNT_1a408}}
Review findings
{{REDACTED_PERSON_9b64a}} reported a duplicate charge. The transaction appeared twice on the October statement.
{{REDACTED_PERSON_73fd2}} reviewed the statement. The reviewer asked {{REDACTED_PERSON_9b64a}} to confirm the date.
The account holder confirmed October 3. {{REDACTED_PERSON_73fd2}} recommended a reversal.
Recommended action
Compare the transaction timestamps and confirm whether a reversal has already been issued.
Illustrative PDF preview: Morgan Ellis and Morgan share one typed PERSON token. Riley Chen and Riley share another PERSON token. Role references such as “the reviewer” and “the account holder” stay readable; they are not identifying values to redact. Morgan’s PERSON and EMAIL tokens share 9b64a to show that the email belongs to Morgan, while keeping the value types distinct. The account has its own identifier. Five-character IDs are shortened for this illustration; production token formatting may differ. This demonstrates entity linkage and coreference; actual resolution depends on configured models, and reconstruction can change layout.
Find it. Transform it.
Human approval is optional. Add a review step to inspect and correct the transformed payload before release, or let it flow without manual approval.
These are visual examples, not live processing results. Supported formats, edition, and configured models determine coverage.
The value is in what you can do with the data. Leave matched sensitive identities out, refine the content moving through AI, and capture what currently goes uncollected for your own pipeline and analysis.
01 · De-identification
Leave the identities out.
Give AI the context it needs without the sensitive values your filters match. Replace names, identifiers, and secrets with placeholders; restore the mapped values in responses. Add human review before release where required.
02 · Transformation engine
Refine what flows through.
Define how data changes as it moves through your AI workflow. Apply configured filters to supported text and attachments, preserve useful context, and inspect the transformed payload before it leaves.
03 · Data pipeline & analytics
Mine what went uncaptured.
Capture AI activity that currently goes uncollected. Retain supported records in your own storage, bring sanitized events together, and build a foundation for mining, workflow analysis, and business insights.
Clean a document batch. Set the rules. Run it across your fleet.
De-identification is a use case of its own. You can scope Transmute into a document-processing workflow without making an AI conversation the end goal.
Documents in. De-identified copies out.
Build a bulk workflow for supported documents: detect sensitive values, apply your transformations, optionally review, and export the processed output for sharing or downstream analysis. For example, prepare a batch of reports with names and account identifiers replaced while retaining the useful content.
Bulk ingestion and document export are scoped implementation workflows; agree supported formats and output fidelity for your deployment.
Your data has its own fingerprints.
Add custom filters for internal account IDs, project codes, proprietary identifiers, and secret formats. Test them against representative examples and put the same rules to work across your workflow.
Custom filter packs: App Pro/Max and supported Platform configurations.
One standard. Managed installs.
Distribute configuration through your MDM or configuration-management tools. Validate it, roll it out, and use supported managed policies to lock organization-controlled settings while leaving permitted choices to users.
Managed policy enforcement requires the Enterprise build. Fleet rollout uses your management tooling.
Bring the evidence together.
Federate supported sanitized audit records from multiple installations into your own S3-compatible storage. Review activity across teams, export evidence, and apply configured retention controls for governance and downstream analysis.
Federation requires a supported Enterprise configuration. Missed sensitive values can remain in sanitized records; protect the destination accordingly.
Capture, audit, and federation capabilities depend on your edition and configuration. Broader mining and intelligence analytics are the direction we’re building toward.
Put it to work
Don’t take the mockup’s word for it.
Start a seven-day free trial. Download without an account, connect a supported tool, and test with fictional data before using sensitive material. Trial eligibility is checked by the licensing service; subscribe and activate with a license key to continue after the trial.
Install the app for your operating system and configure a supported AI tool to route through Transmute. Installing alone does not cover unrelated traffic.
02 · Test
Use a fictional name, email, or test secret that matches an enabled filter. Make a request through the connected tool.
03 · Inspect
Compare the original and transformed content using the inspection features in your edition. Confirm the matched value was replaced and the request still has useful context.
Know what crosses the boundary.
Your tool → Transmute’s configured detection and transformation → optional approval → your AI provider. Matched values become placeholders before forwarding. In reversible mode, mappings stay with your Transmute deployment so supported response tokens can be restored. Unmatched content may still reach the provider.
App licensing egress: Core, Pro, and Max contact our Cloudflare-hosted licensing infrastructure. Trials report a pseudonymous machine ID, edition, OS, version, usage counters, and timing/lease metadata. Paid apps stop trial reporting but check for license renewals. These requests carry connection metadata, not your prompts, attachments, secrets, or reconstruction mappings.
Platform builds: the app licensing and trial-reporting module is excluded. Your configured AI providers, model services, and storage destinations still determine other network traffic. Read the data collection and retention details →
Pick the capabilities you need.
Plan
Start here for
Core
Configured secrets and PII filter packs, tokenizing proxy, and local audit log.
Pro
Core plus custom filters, audit dashboard, evidence export, S3-compatible audit shipping, cloud storage integration, and optional Clearance approval.
Max
Pro plus IQ beta: configured NER and face/media detection, managed fleet policies, and federated governance. ML features require supported models; choosing a plan does not configure them automatically.
Optional Clearance approval requires a supported edition and an enabled policy. Platform builds include Clearance at Compliance and Enterprise, and IQ at Enterprise.
Check the fit before using real data.
Does every file get inspected?
No. Format support, extraction limits, filters, and model configuration determine coverage. Check your deployment’s behavior for unsupported or failed processing; do not assume it blocks every uninspectable request.
Does the free trial include an AI provider?
Your AI provider and its usage charges are separate. The trial covers Transmute access, not third-party model usage.
Are the examples proof of detection accuracy?
No. They use fictional data to explain transformations. Validate your own formats and representative test cases. Human approval is optional and does not guarantee perfect detection.
Need help getting connected?
Use Feedback & support with your operating system, app version, and a sanitized description. Do not include credentials or sensitive documents.
Make AI useful. Keep sensitive data under control.
Start with the app on your machine. Transform matched secrets into placeholders before they reach a model, then restore them in the response. Audit and dashboard capabilities depend on your plan.
Connect your supported tools
One local gateway. Configure integrations for supported LLM tools, Claude Code, Codex, Cursor, DeepSeek, Ollama.
Transparent redaction
Secrets and PII your packs match become placeholders outbound, then rehydrate inbound. Matched values are replaced before forwarding; mapped tokens can be restored in supported responses.
Belt and suspenders
The plugin catches obvious leaks at the agent-loop boundary; the gateway scrubs the wire. Coverage depends on the enabled filters at each layer.
Desktop coverage
Claude Desktop, Cursor, VS Code Copilot, Zed, Continue, routed through the gateway with backup, restore, and health checks per app.
Redacted tokens. Sanitized payloads. Matched PHI never sent in the clear.
What your developer saw
AKIA2X9...REAL_KEY Sarah Johnson eyJhbGci...full_token ...full response
Real values. Full responses. Nothing hidden from the developer.
Platform Capabilities
Take the gloves off your AI agents.
You gave your agent a job, then tied its hands. Let it work. Transmute’s in your corner. You can retire from babysitting.
De-identify documents in bulk
Scope a document-processing workflow that transforms supported files and exports de-identified output for sharing and analysis. Formats, batch ingestion, and output fidelity are agreed during implementation.
Your filters. Your fleet standard.
Define custom filter packs and distribute configuration through your management tools. Enterprise managed policies can enforce organization-controlled settings.
A pipeline shaped around your work
Adapt capture and transformation to your data and workflows through supported extension points.
Know when the pipeline needs attention
See the health of connected services and recover from failures with managed retries and circuit breakers.
Bring the activity together
Federate sanitized events into your own S3 storage. Explore recorded activity across hosts and teams while raw originals stay on their source machines.
Make transformations accountable
Keep a tamper-evident record of transformations. Review what happened and export evidence for your compliance program.
Process sensitive content on your infrastructure
Run configured detection models in your environment. Add contextual analysis without sending that detection work to a vendor-hosted service.
Preserve useful context
Use supported entity detection and linkage to replace sensitive values while retaining the roles and relationships a model needs.
Subscribe to the App. Work with us through a Platform client relationship. Or choose Source when you want a perpetual source license.
Start with a seven-day free trial. Subscribe when you’re ready to continue. Choose Core, Pro, or Max for macOS, Windows, or Linux.
Download without an account. Start a seven-day free trial; activate a paid subscription with your license key to continue.
Subscriptions renew automatically at the same per-seat price and billing interval until cancelled. Cancel anytime by email; see Refunds, Renewals & Cancellation and the Acceptable Use Policy. Purfect Labs sells Transmute; payments are processed by Stripe.
Transmute Core
$49.99/mo per seat$55.54
per seat · up to 6 seats
✓Secrets & PII detection with reversible tokens
✓Supported text, file attachments & PDFs
✓ZIP, TAR & gzip archive processing within configured limits
✓OpenAI-compatible, Anthropic & Gemini API traffic via supported integrations
✓Local tamper-evident audit trail
Agree to the License Agreement and Terms above to subscribe.