{ }tools: 7lang: pljson_schema
Understand the task
Task type, language, context length, and available tools form a request fingerprint. Simple rules are enough to start.
LESS SPEND. MORE COMPLETED TASKS.
Every task.
The right model.
Not every request needs your most expensive model. Route agent tasks to the model most likely to get the job done — within your budget.
THE RIGHT ROUTE01 / 03
Your agentOne request. Multiple possibilities.
APItask: tool_calling · lang: pl
▚
Fabryka RouterClassify. Compare. Route.
✳QQwenSELECTED ✓
MMistralALTERNATIVE
CClaudeALTERNATIVE
↳Verify outcomeschema valid ✓
↺ Every verified outcome informs the next decision.
REQUEST → OUTCOME → LEARNINGMVP CONCEPT
ONE LAYER BETWEEN
YOUR AGENT AND YOUR MODELS
YOUR AGENT AND YOUR MODELS
OpenAI API
OpenMercato
Custom agents
Your models.
Your rules.
Your rules.
01 — FROM REQUEST TO RESULT
Better routing starts
with the outcome.
A plausible answer is only the beginning.
Think valid tool calls, correct JSON,
and tasks that actually get completed.
Choose the best route
Compare each model’s historical success with its cost and latency. Your priorities determine how those trade-offs are weighted.
✓ JSON schema✓ Tool arguments↺ Routing stats
Verify. Learn. Repeat.
A validator checks the response. If it fails, a fallback provides another chance — and evidence for the next routing decision.
02 — A DECISION YOU CAN EXPLAIN
Same request.
Your priorities.
Explore how changing your priority
changes the model. Every choice explained.
EXAMPLE REQUEST ↘
“Find the customer’s order
and update the delivery address.”
Interactive demo · illustrative data.
Not benchmark results or current model pricing.
MODELSUCCESS RATECOST / TASK
Q Qwen 27B91.2%$0,004
M Mistral Small94.8%$0,009
C Claude Sonnet97.1%$0,041
Mistral SmallSELECTED
Balanced: a higher success rate than Qwen at a lower cost than Claude for this type of task.
GET /v1/routing/explainThe cheapest answer ≠ the cheapest completed task.Account for retries, fallbacks, and tool execution. ↗
03 — A SMALL CHANGE IN YOUR CODE
Your agent.
Smarter decisions.
A familiar API with a routing layer underneath. Choose automatic selection or set a specific priority for each request.
fabryka/autofabryka/cheapfabryka/qualityfabryka/fastProposed Smart Router API. These model aliases and the Responses example describe the MVP design. For the current API, see the documentation.
quickstart.py
import os
from openai import OpenAI
client = OpenAI(
base_url=os.environ["FABRYKA_BASE_URL"],
api_key=os.environ["FABRYKA_API_KEY"]
)
response = client.responses.create(
model="fabryka/auto",
input="Find the customer’s order…"
)START WITH YOUR WORKLOAD
Stop guessing which model.
Find the one that delivers.
Compare models on the same tasks. Set your quality threshold.
Then route traffic based on the evidence.