Eagle // Logic
Product Showcase Pricing Support

PRIOR Product Documentation

Welcome to the PRIOR product documentation. This is the customer-facing guide set for operators and integrators running a licensed PRIOR engine (v1.2.2).

PRIOR is a licensed, self-hosted inference engine that enforces behavioral and factual policy by injecting deterministic constraints directly into a model's latent space, at a calibrated golden layer in the residual stream. Because policy is applied geometrically rather than through prompt text, it holds up against obfuscation, jailbreaks, and stale context. PRIOR speaks the standard OpenAI chat API, so it drops in behind any existing client.

© Eagle Logic. Proprietary and licensed. Not open source. Use requires a valid, node-locked license. See Licensing.


Start here

If you want to… Read
Run the engine and make your first request Getting Started
Understand how PRIOR enforces policy Concepts
Integrate against the HTTP API API Reference
Deploy to production (Docker/Kubernetes) Deployment
Author or ingest your own policies Policy & Packs
Operate from a browser dashboard WebUI Operator Guide
Understand tiers, node-locking, capabilities Licensing

Need a link you can share? These docs are public, so any page here is safe to send on. If you're pitching a colleague or a security reviewer, the measured results and the interactive demos on the website make the better link.


The 60-second picture

                       ┌──────────────────── PRIOR engine (one container) ─────────────────────┐
   OpenAI-compatible   │                                                                        │
   client ───Bearer───▶│  POST /v1/chat/completions                                             │
                       │     │                                                                  │
                       │     ├─▶ Tri-State WHEN gate  ── routes every prompt BEFORE generation  │
                       │     │      🟢 GREEN  → answer normally (bit-for-bit vanilla)            │
                       │     │      🟡 YELLOW → answer + inline confirm notice                   │
                       │     │      🔴 RED    → deterministic refusal via golden-layer steering  │
                       │     │                                                                  │
                       │     └─▶ x-prior-* response headers (zone, pack, decision id)            │
                       │                                                                        │
   operator ──Bearer──▶│  /admin/*   governance: packs, keys, license, calibration, audit       │
                       │  /metrics   Prometheus   ·   /health   liveness                         │
                       └────────────────────────────────────────────────────────────────────────┘
  • One immutable container, no external calls in the inference path (air-gapped tier available).
  • Native Rust serving path with no Python in production. Default port 8089.
  • OpenAI-compatible. A vanilla client that never sends PRIOR's prior_options extension is completely unaffected; policy still applies.
  • Operator-owned policy. You ship, tune, mute, or author detection packs. They hot-reload with no restart.

Product surfaces

Surface What it is Port (default)
Engine API The OpenAI-compatible inference + /admin governance server 8089
WebUI An optional browser dashboard (a standard API client with no special privileges) 8080
License A node-locked, ED25519-signed license.bin from Eagle Logic n/a

Conventions used in these docs

  • Bearer <key>: an API key issued through /admin/keys (or seeded via PRIOR_API_KEY).
  • 🟢 / 🟡 / 🔴: the three gate zones (GREEN / YELLOW / RED).
  • prior_options: the optional vendor-extension object on chat requests.
  • Roles are Observer / Operator / Admin.

Questions or licensing: support@eagle-logic.com · https://eagle-logic.com.