Run AI inside your own walls.
Prove nothing leaves.
Precepta is a self-hosted, governed gateway for open-source and hosted models. Every request is policy-checked, PII-redacted, and audited — and a cryptographic attestation proves your data never crossed the boundary.
Most companies can't safely deploy AI.
Prompts and data leave the building for third-party APIs. There's no consistent governance across teams. And when an auditor asks what happened to a request, there's no answer. Precepta closes all three gaps — without changing how your developers work.
One governed path for every request
The rails are fixed; the routing is yours. Every call flows through the same governed pipeline before — and after — it reaches a model.
Governed pipeline
Firewall, sensitivity, policy, PII redaction and an output leak scan run on every request — before a model ever sees it.
Smart Router
Set model:"auto" and the router picks the best healthy in-boundary model per request — quality vs. cost vs. latency.
Traces & audit
Replay the full governed journey of any request — every step, the decision, and why — with a tamper-evident audit log.
Policies & budgets
Per-key scopes, cost and token caps, and policy rules enforced before inference — no surprise spend, no ungoverned access.
Cache & compression
Reuse deterministic answers and trim long prompts. Real, per-endpoint savings you can see — never at the cost of governance.
Sovereignty attestation
A live egress probe and a signed attestation prove — not promise — that nothing left your boundary.
Not a setting you trust. A property you can prove.
Precepta runs on an internal network with no route to the public internet. Reaching an external host is impossible unless you explicitly, owner-only, approve it — and the attestation records exactly what your posture is.
- ✓Zero-egress by construction The app has no network path out — a live probe confirms it at boot and on demand.
- ✓Owner-gated approved egress Reach a specific hosted model only if the platform owner allows that exact host. Everything else stays sealed.
- ✓Signed attestation + full audit Show auditors a cryptographic record of your sovereignty posture and every governed request.
# Same OpenAI SDK — only the base URL changes.
from openai import OpenAI
- client = OpenAI()
+ client = OpenAI(base_url="https://ai.yourco.com/v1")
resp = client.chat.completions.create(
model="auto", # "auto" = Precepta Smart Router
messages=[{"role": "user", "content": "hello"}],
)
print(resp.choices[0].message.content)The API they already know. Change one line.
Precepta speaks the OpenAI standard — chat, embeddings, moderations, models. Point any OpenAI SDK at your gateway and every call is governed, routed and audited. No client rewrite, no lock-in.
Deploy AI you can defend.
Stand up a governed, sovereign AI gateway on your own infrastructure — free for 15 days. Your models, your data, your walls.