TASK-AWARE
The route follows the work, not a fixed model preference.
AI ROUTING / ONE ENDPOINT
One request in.
The right route out.
Send every request through one OpenAI-compatible endpoint. ELTEX reads the job, checks your routing policy, and picks a capable route—so your product can move fast without hard-coding every provider.
01One endpoint
02Task-aware routes
03Visible decisions
LIVE ROUTING FABRICREADY
One requestPOST /v1/chat/completions
Routine workfast + efficient route
Deep reasoningpremium reasoning route
Tool + code workcoding-optimized route
Long contextcontext-optimized route
CAPACITY-AWARE
Use the right level of intelligence without wasting premium capacity.
FALLBACK-READY
Keep availability and recovery part of the same operating layer.
THE ROUTER LAYER
Your app speaks once.
ELTEX handles the route.
An AI router sits between your application and the model providers. It turns one stable integration into a flexible decision layer: classify the task, choose an eligible route, and keep the outcome visible.
One interface.
Many decisions.
- REQUEST
POST /v1/chat/completions - SIGNAL
task · complexity · context - POLICY
smart · preferred · force - RESULT
model · usage · request_id
HOW THE DECISION HAPPENS
Three moves from intent to output.
Routing should feel invisible to the end user and legible to the team operating the system.
01 / UNDERSTAND→
Read the work, not just the prompt.
ELTEX looks at task type, complexity, context, tools, and the quality bar before a route is selected.
task · complexity · context02 / MATCH→
Match the job to a route.
Routine work can stay fast and efficient. Planning, coding, and synthesis can move to a stronger route when the task calls for it.
model · provider · capacity03 / STAY IN CONTROL✓
Keep the decision observable.
The selected model, usage, request ID, and routing mode stay available for debugging, reporting, and better policy decisions.
route · usage · request IDROUTING MODES
Automatic by default.
Explicit when it matters.
Start with Smart Routing, then take as much control as the workload needs. Every mode uses the same application-facing endpoint.
Smart
Let ELTEX classify the request and select an eligible route based on task, complexity, availability, and policy.
Preferred
Set a preferred model as a capability and cost ceiling while ELTEX keeps the request within that boundary.
Force
Call the exact model you choose when a workflow needs a specific capability, behavior, or evaluation target.
Available on every Agent Plan and active Token Sachet.BUILT FOR REAL WORK
Different workloads.
One routing layer.
Give each kind of work a sensible default without making your application maintain a maze of provider-specific integrations.
Coding & tools
Send tool-heavy steps, debugging, and implementation work to routes optimized for code and structured output.
Research & context
Keep long documents and multi-step synthesis on routes with the context and reasoning headroom they need.
Support & everyday AI
Handle summaries, rewriting, classification, and routine Q&A on efficient routes that keep responses moving.
Agents in motion
Let one agent move between models as it plans, calls tools, verifies work, and recovers from a failed route.
MAKE ROUTING PART OF YOUR STACK
Give every request a better path.
Start with Agent Plan for recurring workloads, then use the live model catalog and API docs to shape the route your product needs.