Skip to main content
By Purfect AI · Runs on your network

Your data doesn't leave your network.

Purfect Shield is a transparent redaction gateway that sits in front of every LLM tool your team uses. Secrets scrubbed outbound, rehydrated inbound. The model never sees what shouldn't leave your environment.

No hosted service. No license server. No subscription. You own the source.

Purfect Shield — Product Demo

Two layers. Two fail-safes. A developer never knows redaction happened — they write the same code, paste the same context, hit the same enter key.

The leak surface

Your developers use Claude Code, Codex, Cursor, DeepSeek — every new agent that ships. And every day someone pastes production data into a prompt by accident. Compliance officers know it. CSOs lose sleep over it.

Cloud credentials
AWS_ACCESS_KEY_ID, GCP service-account JSON
Source-control tokens
ghp_..., GitLab PATs, gho_...
API keys
sk-ant-..., sk-..., internal service tokens
Environment secrets
.env contents read into context by a tool
Database credentials
Postgres / Mongo connection strings
JWTs
eyJ... tokens in headers, args, fixture files
Domain PII
SSNs, patient names, DOBs, card PANs
Tool exfiltration
Tool trying to read ~/.ssh/, ~/.aws/, /etc/shadow

Asking your team to manually sanitize every prompt is asking them to stop shipping.

The Engagement Week

Foundation deploys in 5 days. Not months, not quarters — five working days from kickoff to production-ready.

Day 1
Threat model workshop

We sit with your team, map your actual attack surface — your prompts, your tools, your data. Select your domain filter pack.

Day 2
Gateway deployed

Go gateway live on localhost:9090. BASE_URL wired into your dev shell rcs. Every LLM call now inspectable.

Day 3
Plugin installed

Auto-installed for your chosen LLM CLI under ~/.claude/skills/ or ~/.codex/plugins/. Hooks fire at UserPromptSubmit, PreToolUse, PostToolUse.

Day 4
Eval suite green

30+ client-specific test cases delivered, wired to playground shield eval, all passing. Any future regression caught before it ships.

Day 5
Handover

Runbook, source delivery, 30-day support clock starts. You own the code forever — no license server, no kill switch.

What the model saw
[REDACTED:AWS_KEY_001]
[REDACTED:PHI_NAME_001]
[REDACTED:JWT_001]
...sanitized payload

Redacted tokens. Sanitized payloads. Zero PHI.

What your developer saw
AKIA2X9...REAL_KEY
Sarah Johnson
eyJhbGci...full_token
...full response

Real values. Full responses. Nothing hidden from the developer.

Belt and suspenders — plugin catches obvious leaks at the agent-loop boundary, gateway scrubs the wire. If one layer misses, the other catches.

Three Tiers, One Standard

All tiers include source delivery. No subscription, no license server, no kill switch. Upgrade between tiers at any time — prior engagement credited.

🚦
Foundation
$10,000 · 1 day

One LLM provider. One day. You own the source.

Gateway deployed on your infra, belt-and-suspenders plugin installed, threat model workshop, one domain filter pack, 30+ eval cases, runbook, and 30-day support. Source delivered at handoff — yours forever.

→Gateway (localhost:9090) deployed on your infra
→Plugin auto-installed for 1 LLM CLI (Claude Code, Codex, Cursor, DeepSeek, Ollama)
→Threat model workshop — Day 1, with your team
→Default filter packs: generic secrets, AWS credentials, GitHub tokens, common PII
→1 domain filter pack — healthcare, finserv, legal, education, or ecommerce
→30+ client-specific eval cases, all green
→Runbook + full Go source delivery
→30-day email support
Most Popular
📋
Compliance
$25,000 · 1 week

Foundation + the audit trail your compliance officer signs off on.

Everything in Foundation, plus a local audit sidecar and dashboard. Every redacted call becomes evidence — who triggered it, what CLI, what was redacted (categories only, never raw values), when. SOC 2 and HIPAA evidence ready to export.

→Everything in Foundation
→Local audit sidecar — append-only JSONL, tamper-evident hash chain
→Dashboard on localhost:9091 — real-time redaction stream, filter by user/date/model
→Audit schema documentation (stable contract for your auditors)
→SOC 2 / HIPAA evidence export — one command
→Log retention runbook (rotation, compression, archival)
→90-minute compliance team training
→S3 Federation upgrade available at +$8,000
🌐
Enterprise
$45,000 · 3 weeks

Federated audit across every host. Custom filter packs. The full deployment.

Compliance across your entire org — dev workstations, CI runners, production sidecars — all aggregating to a single S3 bucket. Up to 5 custom filter packs, filter authoring SDK, CI eval gates, on-site training, and 90-day support.

→Everything in Compliance
→S3 federation — every host ships events to your bucket, your keys
→Federated dashboard — org-wide view, drill by host / user / team
→Up to 5 custom filter packs authored to your domain
→Filter authoring SDK — your team extends Shield without us
→CI eval gates wired into your build pipelines
→Semantic Sentiment Filter — on-prem ONNX, 5-axis intent scoring (included free)
→2-day on-site training (engineering + compliance teams)
→90-day support + 2 office-hours sessions

Add-ons

Available on any tier. Semantic Sentiment Filter is included free at Enterprise.

Additional LLM provider$2,000
Additional filter pack$3,000
Semantic Sentiment Filter (on-prem ONNX, 5-axis)$5,000
S3 Federation upgrade (Compliance → Federated)$8,000
CI eval gate — single pipeline$3,000
CI eval gate — multi-pipeline / multi-repo$5,000
Retainer (async + office hours)$5,000/mo
Annual refresh (source + filter pack updates)$1,000/yr
🔒

Your data never touches our infrastructure

The gateway runs on your network. The audit log writes to your S3 with your encryption keys under your retention policy. There is no Purfect Shield cloud. There is no telemetry. Audit teams query their existing pipeline.

SOC 2 evidence readyHIPAA audit trailYour S3 + your keysAppend-only logNo license serverSource code delivered
Intake Portal

Let's Build.

Submit your technical details and we will formulate a production scope, architectural dependencies, and exact model selection profiles.

48-Hour Response SLA
Every request is routed directly to a principal systems engineer.
Zero Cloud Risk
All contracts guarantee strict IP ownership and security boundary isolation.

Prefer a call instead?