SpiderGate routes 32 task aliases over 91 fallback slots

TALLINN, Estonia, August 10, 2026 — SpiderGate today detailed the routing layer behind its OpenAI-compatible LLM gateway at https://spideriq.ai/api/gate/v1, which lets developers address a job rather than a model so that agents survive a provider retiring an id. The gateway serves 32 task aliases backed by 91 ordered fallback slots, an average of 2.8 models deep, in front of a catalog of 1,216 models of which 61 are key-backed [source: https://spideriq.ai/api/gate/v1/aliases].

The change targets an ordinary and expensive coupling. An agent naming a specific model id inherits its provider's product decisions: when that id is retired, throttled or repriced, every service that named it fails at once and each needs a deploy to recover.

Under SpiderGate a caller sends a task name such as spideriq/coding in the model field. The gateway resolves it to a chain, reorders that chain by live provider health so models on struggling providers move back rather than being removed, and serves the first candidate that answers, per the documented resolution order at https://gate.spideriq.ai/docs/using-the-gateway/task-aliases. If a chain is exhausted the request falls through to its family's general alias instead of failing. When the model that answered is not the first choice, the response carries an X-SpiderGate-Fallback-From header naming it, per https://gate.spideriq.ai/docs/reference/api.

"The measurement that convinced me was an accident," said Martin Shein, founder of SpiderGate. "I checked our own alias list against the database while writing the documentation for it, and the documentation was wrong: it said 26 aliases when there were 32, and 16 of the 26 rows it listed named a model that no longer led that chain. Sixteen routing changes had happened underneath and not one caller had to change a line, because no caller had named a model. The docs rotted and the callers did not."

The aliases sit in three families by traffic type: 18 spideriq/* for batch workers where cost dominates, 9 agent/* for live multi-turn sessions, and 5 opvs/* for the OPVS agent platform, which splits by job size, per the alias endpoint [source: https://spideriq.ai/api/gate/v1/aliases]. Chains route across Cerebras, Groq, Mistral, MiniMax, NVIDIA NIM, OpenRouter and Codex.

"An alias is not a recommendation engine and we are careful not to describe it as one," Shein said. "It does not read your prompt and pick a model per request. It resolves a name to an ordered list we maintain, and reorders that list by whether a provider is currently answering."

The live list, with each chain in priority order and its 30-day usage, is published at GET https://spideriq.ai/api/gate/v1/aliases and needs no authentication. Routing is executed by BerriAI's litellm.Router, which performs least-busy selection, retries and provider cooldown; SpiderGate adds the task alias, per-tenant authentication and a usage record per call. Documentation: https://gate.spideriq.ai/docs/using-the-gateway/task-aliases

About SpiderGate

SpiderGate is the LLM gateway built for agencies: keys live encrypted in a Vault you contribute to without exposing them, and a shared pool with least-busy failover keeps requests flowing when a key rate-limits or dies. On top sits Studio — a chat workbench where you switch between 1,200+ models mid-conversation, with your files, notes, and saved prompts organized by project.

Media Contact

Press Office Media Relations, SpiderGate press@spideriq.ai