Three surfaces. One redaction engine. Six ways to wire your rules. And a dashboard that — politely — turns AI from a leaky tap into a team-wide skill.
Read time: about 8 minutes. Skip-the-diagrams time: 3. Show this to your CISO time: 30 seconds.
Most security tools punish you for working. You write a prompt with an API key in it, the tool blocks you, you context-switch, you lose the thought. SecureAIFlow does the opposite — it lets you keep typing the way you already do, intercepts the dangerous bits before they leave your IDE, and quietly puts your real values back into the answer.
You never see [REDACTED_1] in your editor. The LLM never sees your secret. Both halves of that sentence matter.
Early versions of SAF showed the redaction in the IDE. Devs hated it. They'd see [REDACTED_1] in a code example from the model, paste it into a file, run it — and break things. They started turning the extension off.
So we reversed the model. The IDE sees clean, runnable code. The audit log sees every redaction. The LLM sees nothing it shouldn't. Three different views of the same prompt, each tailored to what the viewer actually needs.
Multi-turn conversations with model selection — Claude Sonnet, GPT-5, Gemini Flash. History persisted per workspace; fully redacted in transit.
Select code, right-click, ask. Inline diffs with native VS Code Accept / Reject CodeLens — no custom UI to learn.
Fast tier (Gemini Flash Lite, Haiku) for short prompts. Power tier (Sonnet, GPT-5) for the gnarly stuff. Waterfall fallback when a provider has a bad day.
Bring your own keys and we stay out of the billing relationship. Or use ours and get one bill, smart routing, and cost caps you can actually enforce.
Native Markdown ingestion. Chunk API for everything else. Ground answers in your own docs — without uploading them to a public model.
OpenID Connect for team auth. Policies inherited from the central dashboard. The dev configures nothing. The DSI is happy. Everyone wins.
Your devs live in their IDE. Your analysts live in a browser tab. Your marketing lead just wants a chat that doesn't ask her to install anything. They're all using AI. They all need protection. They do not all need the same UX.
For people already living inside ChatGPT all day. We don't ask them to switch tools. We ride along — invisible until a secret tries to leave.
For everyone else. The PM doing strategy work. The analyst summarizing a deck. The HR lead writing a job spec. Open the URL, start chatting.
For developers who already have an AI workflow and don't want a new one. Install from the Marketplace; coexists peacefully with Copilot and Cursor.
A senior backend dev and a customer support agent need different things from the same security layer. We stopped pretending otherwise.
SecureAIFlow catches two things: secrets (API keys, passwords, tokens, the stuff in your .env file) and PII (the customer data sitting in your database — emails, IBANs, internal IDs). They live in different places at your company, so we read them from different places too.
Banks already have HashiCorp Vault for secrets. Fintechs already have Postgres or Mongo for customer schemas. We didn't want to become one more database your security team has to migrate to — so we connect to wherever you already keep the truth.
.yaml files, or matched by regex.Point SAF at your customer schema. Tell it which columns are sensitive — email, iban, phone, that custom customer_ref. SAF looks up patterns at runtime — your data never moves.
Same idea, document-shaped. Flag the fields that hold PII in your collections; SAF reads the schema and detects matches in prompts. Change streams mean rule updates propagate in seconds, not minutes.
The gold standard. Every API key, password, and token your platform team has already onboarded into Vault is automatically protected — SAF reads the secret signatures, never the values. 100% precision on anything Vault manages.
Drop a file in the config directory, commit it, you're done. GitOps for security policy — versioned, reviewed, rolled back the same way you handle Terraform. Honest workflow for small teams.
Some patterns nobody hard-codes for you. Email is the classic — here's the one we ship as a starter:
# email [\w.+-]+@[\w-]+\.[\w.-]+ # French IBAN FR\d{2}(\s?\d{4}){5}\s?\d{3}
Add your own for internal IDs, SKUs, or that weird prod token format only you have.
REST endpoints for rules, integrations, audit logs, dashboard exports. Wire SAF into your SIEM, your CI/CD, or the internal admin tool nobody else maintains.
The CISO needs an audit trail. The engineering manager needs to know who on the team is actually shipping faster because of AI — and who's quietly stuck. Same data. Two different questions. Same screen.
Most security dashboards stop where ours starts. Yes, you'll see secrets caught, tokens scanned, cost per provider, redactions per surface — all the standard audit fare. Your DPO will be fine.
But scroll down. You'll find the team view: consumption per person, models used, prompts sent, redactions triggered, cost incurred. Sort by activity and the people pulling the most leverage out of AI rise to the top. They're your top performers. Not because they're "10× developers" — because they figured out a few patterns the others haven't found yet.
You also get cost-per-team, cost-per-project (use tags), DORA Article 28 audit exports, and the full prompt history — redacted, naturally — for any compliance investigation.
Free for individuals. Fair for teams. Hosted in Paris or Frankfurt — never anywhere else. Your DPO will thank you. Your devs will not notice we're there. Which is exactly the point.