Input & validation
Normalize requests, reject malformed input and apply size/permission limits before any model sees the work.
ARIDON MODEL ROUTER
One Aridon. The best available AI for each job.
The user should not have to decide whether a task belongs in ChatGPT, Claude, Gemini, Grok, DeepSeek or a local model. Aridon classifies the work, routes it, preserves company context and falls back automatically.
PROVIDER POOL
Use the specialists that are actually connected.
ROUTING TABLE
What Aridon does with each kind of work.
| Task | First choice | Fallback |
|---|
FULL AGENT OPERATING LOOP
The architecture behind the button.
Task classifier
Detect research, social intelligence, coding, multilingual, long-context, creative, private-local or general executive work.
Best-model router
Send the job to the strongest configured provider for that task and fall back automatically if it fails.
Company Brain & memory
Pull tenant context, company knowledge, uploaded files, executive memory and semantic retrieval into the task.
Supervisor loop
Plan, delegate, inspect specialist output, retry weak work and stop when the definition of done is met.
Approval-gated execution
Only approved tools can act. Consequential actions remain behind human approval gates.
Observability
Track provider attempts, latency, fallbacks, usage events, audit records and final outcomes.
Safe output
Return the answer, file, deck, report or queued action with the routing and approval state preserved.
PRIVACY / LOCAL MODE
Private jobs can stay on a local model.
If an OpenAI-compatible local endpoint is configured, requests that explicitly ask to remain local or offline can be routed there first. That supports Ollama, LM Studio, vLLM and similar local deployments without changing the Aridon interface.