8B and 14B fast lanes
Use 8B for instant classification, language cleanup, policy tags, and simple traveler answers. Step up to 14B for lightweight trip edits, memory lookup, action drafting, and low-risk support tasks.
Private model fabric for apps
Build calm, capable travel apps with Enjway AI.
A model fabric for itinerary reasoning, route repair, trip memory, and production-ready user actions. Designed to feel simple at the surface and rigorous underneath.
Research
A model ladder with clear jobs.
Keep fast work fast. Send hard reasoning to deeper routes. Measure every path before it reaches a traveler.
30B and 32B operating tier
Route shopping, itinerary repair, margin checks, inventory comparison, and order-service decisions through 30B/32B models when the workflow needs context but still has to stay quick.
70B deep reasoning route
Reserve 70B for complex multi-city planning, disruption recovery, refund edge cases, direct-connect tradeoffs, and operator-reviewed decisions where accuracy matters more than latency.
Products
Intelligence that shows up as useful product.
The interface stays quiet: route suggestions, itinerary edits, clear confirmations, and a review trail your team can trust.
Trip Copilot
Live
Weekend plan
Bengaluru without the rush
8 stops, 2 calm dinner windows, 1 weather-safe backup
deep-travel-reasoning
Premium itinerary optimization needs deeper reasoning and stricter review.
Developers
An API surface your team already understands.
Use familiar chat completions, streaming responses, model names, and structured actions while Enjway handles product-specific policy.
POST /v1/chat/completions
{
"model": "enjway-trip-copilot",
"messages": [
{ "role": "user", "content": "Repair my Sunday route" }
],
"response_format": { "type": "trip_action_contract" }
}Product fabric
Private LLM infrastructure for Enjway's app intelligence.
The app keeps a simple Copilot surface while the backend handles model selection, signed action contracts, quality scoring, repair passes, and rollout gates across 8B, 14B, 30B, 32B, and 70B routes.
OpenAI-compatible, private by default
Enjway talks to a backend-only `/v1` gateway, so the mobile app never receives model keys. vLLM-backed routes can serve small fast models and deep reasoning models behind one API shape.
- Runtime
- Self-hosted gateway, bearer token, `/v1/chat/completions`
- Catalog
- 8B, 14B, 30B, 32B, and 70B model lanes
Send each request to the right tier
The router chooses frame, default, repair, deep, or frontier paths from context size, user tier, expected cost, latency budget, health status, and whether the result needs review.
- Fast path
- 8B/14B for frame extraction and safe low-risk edits
- Deep path
- 32B repair and 70B premium reasoning when quality matters
Ship model changes through evidence
Every route can be gated by smoke checks, JSON validity, judge score, repair rate, fallback rate, p95 latency, canary promotion, and ops audit trails before broad rollout.
- Contracts
- Typed actions, signed selections, structured responses
- Ops
- Quality dashboard, audit events, canary and repair controls