GhostRouter Read Docs
AI‑native trust • zero‑trust identity • smart routing

Route every AI request to the right lane

GhostRouter is an AI‑native identity + routing layer that enforces privacy and accountability, then selects Local / Cloud / Hybrid / WebNav using trust, cost, latency and telemetry

Try Demo See how it works GitHub
Full JSONL audit trail
Privacy budgets + lane allow/deny
Persona/PersonalToken model
Verify‑mode when risk is high

Why GhostRouter?

Most AI stacks treat every request the same. GhostRouter adds a trust layer in front of your models

Zero‑trust by default

Identity is explicit (tokenized), permissions are bounded, and risky flows can be sandboxed or denied.

✓
Privacy first: keep sensitive data local unless policy allows.
✓
Accountability: every decision leaves a verifiable trail.

Routing is intelligence

Models are tools. GhostRouter decides which tool is safe, fast, and cost‑effective for each request.

✓
Cost control: budgets + lane caps prevent surprise bills.
✓
Reliability: fallback + verify‑mode when uncertainty rises.

Core features

The minimum set you need for production routing: identity, policy, telemetry, audit and executors.

PersonalToken / PersonaToken

Cryptographically verifiable identity that binds the agent to persona, scopes and permissions.

Trust Engine

Tracks success/failure + telemetry signals to compute trust and drive policy decisions.

Smart routing

Scores lanes by trust, privacy risk, cost ceiling and latency targets — with safe fallbacks.

Privacy guardrails

Allow/deny lanes, strict local‑only modes, redaction hooks, and critical‑domain gates.

Telemetry

Entropy, drift, latency, device pressure — used as an early‑warning system (not direct control data).

Hybrid executor

Parallel brainstorming + verification: consolidate multiple outputs into a stable answer.

Routing lanes

A simple mental model: four lanes, one policy engine.

Local

Fast, private inference on your own GPU.

default for low‑risk

Cloud

Higher quality/reasoning when budgets + policy allow.

gated by trust & privacy

Hybrid

Use multiple models in parallel, then verify and stabilize.

best for “hard” tasks

WebNav / Honeypot

Verification via web tools, or sandbox suspicious workflows.

evidence‑first

How it works

Three stages, one outcome: safe execution with a trace.

1
Identity Engine validates token, scopes, expiry, and risk posture.
2
Telemetry Engine enriches signals (entropy/drift/latency/pressure) and applies hysteresis.
3
Route Engine scores lanes, applies overrides, dispatches to executor, and logs JSONL.

Design rule

Telemetry is not direct control data. It is an early‑warning system that triggers verification, escalation, or stricter policy checks. 

# pip install ghostrouter  (example placeholder)
from ghostrouter import route_request

result = route_request(
    prompt="Can I take ibuprofen with my blood thinner?",
    persona_token="persona:gary-lab",
    budgets={"max_cost_usd": 0.03, "max_latency_ms_p95": 900},
    policy="critical-domain-guard",
)

print(result.lane, result.decision_reason)

Docs & links

Hook these to your real endpoints when you deploy.

Read Docs GitHub Contact Try Demo

Demo story (placeholder)

When it matters, GhostRouter forces verification. Replace this block with your real demo flow.