For developers

nAIvigate data, in your agents and your code

The same engine as the site: Finder, cost, Hardware Match, Trust Index, changes. Two access methods, one source of truth.

MCP (Claude Desktop, Cursor, ChatGPT…)

Add the server and ask in plain language: “which open-weight model under 24 GB for French RAG?”.

{
  "mcpServers": {
    "naivigate": {
      "url": "https://naivigate.com/mcp",
      "headers": { "x-api-key": "nv_free_…" }
    }
  }
}

REST JSON

The same tools over HTTP, for scripts and integrations.

curl -H "x-api-key: nv_free_…" \
  "https://naivigate.com/api/v1/recommend?who=me&use=code&where=local&machine=gpu24"

Available tools

recommend_model

The Finder: 3 models that pass your constraints, with reasons

GET /api/v1/recommend?who=org&use=rag&data=internal&keep=eu

models_for_hardware

Which models run on a configuration

GET /api/v1/hardware/rtx-4090-24-go

estimate_cost

Monthly cost, API vs self-hosted

GET /api/v1/cost/claude-sonnet-5?tokens_millions=20

get_trust_profile

Trust Index and compliance of a model

GET /api/v1/trust/gpt-6-astra

get_model_changes

What changed (price, Elo, licence, hosting)

GET /api/v1/changes?days=7

search_models · get_model · compare_models · cheapest_for · get_podium

Catalogue, cards, comparison, podium

GET /api/v1/models/claude-sonnet-5

Access

Free0 €

60 requests / minute without a key, or 1,000 / day with a free key. “Source: nAIvigate” attribution required in what you publish.

Profrom €49 / month

50,000 requests / day, full responses (20 Trust Index criteria, provenance of every field, compliance fields), no attribution required. For product integrations and consultancies.

Enterpriseon quote

Custom quotas, data licence, full export with history, support. On quote.

Rules

Read-only. Verified, dated data, never guessed. No position in our rankings is for sale. Prices are in US dollars per million tokens, as published by vendors.

MCP (Claude Desktop, Cursor, ChatGPT…)

REST JSON

Available tools

Access

Request a key

Rules