For developers · OpenAI & Anthropic compatible · Self-hostable

Change one line.
Every AI call becomes provable.

Point the SDK you already use — OpenAI, Anthropic, or Claude Code itself — at the Pulse Gateway. Personal data is masked before egress, the right model is picked per call, and every completion ships a signed certificate an auditor can verify offline.

Drop in — 5 minutes▸ Watch the 2-minute demo
No SDK to learnStreaming & tools pass throughRuns on your infrastructure
const client = new OpenAI({
-  apiKey: process.env.OPENAI_API_KEY,
+  baseURL: 'https://api.pulse-labs.fr/v1', apiKey: process.env.PULSE_API_KEY,
})
// response headers:
x-pulse-proof: pf_a059713a…  // signed · chained · offline-verifiable
x-pulse-pii-masked: 2
The shadow-AI problem

Your team already ships prompts to third-party clouds.
Nobody can say what was in them.

Your AI traffic already leaks

Prompts carry customer names, emails from fixtures, snippets of production data — straight to third-party clouds, with no register of what left or when.

Nothing is provable afterwards

When compliance, a customer or an auditor asks what your AI did and on what basis, logs you wrote yourself prove nothing. You have no opposable record.

Rewriting apps is not an option

Every fix that requires touching application code dies in the backlog. The only integration budget that always exists: one config line.

What happens on every call

A pipeline you don't write,
on traffic you don't rewrite.

1

PII Shield

Names, emails, IBANs, phone numbers are swapped for consistent neutral tokens before anything leaves — and restored in the answer. Medical mode covers HIPAA Safe Harbor identifiers.

2

Model routing

Cheap model for simple asks, full model for complex work and agent loops, local model when forced sensitive. You never hardcode a provider again.

3

Capabilities by address

The model field is an address: agent:cfo runs your workspace agent, +cortex grounds the call in your knowledge base, +sensitive forces local. Composable, opt-in, never imposed.

4

Signed certificate

Every completion appends an Ed25519-signed, hash-chained attestation: request, redacted egress, grounding sources and response — committed by sha256, never by content. Verifiable offline.

// The model field is an address — capabilities are one string away:
"pulse-auto"  · routed per call   "pulse-sensitive" · forced local, cert proves it
"agent:cfo"  · your workspace agent answers (memory, docs, tools)
"pulse-auto+cortex" · grounded in your knowledge base, sources hashed into the cert
"agent:cfo+nomemory" · inherits nothing, leaves nothing — flags subtract too
Works with your stack

Anything that speaks OpenAI or Anthropic speaks Pulse.

OpenAI SDKAnthropic SDKClaude CodeLangChainCrewAIn8n / MakeCursor / Continuecurl

Claude Code runs entirely through the gateway — every turn of the agent loop masked, routed and attested — with two environment variables. Agent frameworks and automation tools plug in through their standard base-URL setting.

Sovereign by construction

A cloud proxy can promise.
Inside your walls, it can prove.

Self-host the whole thing

The gateway ships as a standalone service (Docker or bare Node). Run it on your own infrastructure — the same binary that powers our cloud.

Your keys, your registry

Each install generates its own Ed25519 signing pair. Your proof chain is yours: nobody — including us — can mint or alter a certificate for your traffic.

Local models included

Route sensitive traffic to Ollama on your own hardware. The certificate then proves the data never left the machine — cryptographically, not contractually.

Get started

Point your SDK at the Gateway.

Five minutes from an API key to a signed, offline-verifiable certificate — with the SDK you already have installed.

Open the quick startRead the gateway reference