Model-agnostic AI agents: OpenAI, Claude, Gemini and a local model in one platform

Alexander Krastev · 2026-10-19

The AI model you pick today will not be the best or cheapest one in six months. An automation platform that ties every agent to one provider makes that your problem. Model-agnostic AI agents make it a setting.

One platform, many providers — and your own

AIK talks to the big providers — OpenAI, Anthropic (Claude), Google (Gemini), Groq and others with an OpenAI-compatible interface — and to local models running on your own machine through Ollama. The model list is checked against each provider regularly, so a retired model does not silently break an agent.

Chosen per agent, not per company

Each agent has its own model setting:

• Local — a model on your server. Nothing leaves the building, no per-call fee. Right for invoices, payroll, anything private.
• Cloud — the strongest or cheapest cloud model for the job.
• Smart — simple tasks go to a small, cheap model and hard ones to a strong model, by a cascade you can change.
• Auto — the account's default.

So the agent reading supplier invoices can stay local while the one drafting a LinkedIn post uses a cloud model.

Your keys, your bill

You bring your own API keys; nobody marks up the model price. Every run is priced on the model it actually used and shown in the run list, and a budget per agent stops spending at the limit.

When a provider has a bad day

A backup model can be named: if the provider's service fails, the step retries with the backup instead of failing the whole run.

Try models on your own documents

Pick the model with evidence, not marketing: run two or three models on the same invoices or documents and compare their results and cost side by side before choosing.

More: local LLM AI agents · run AI agents on your own server with Ollama · what an AI agent costs to run.