Skip to content

07_Developer_Access.md

Latest commit

 

History

History
112 lines (80 loc) · 3.02 KB

07_Developer_Access.md

File metadata and controls

112 lines (80 loc) · 3.02 KB
Raw

Developer Access & Integration

BaseModel is designed to be consumed by SDKs, CLIs, agents, dashboards, and other applications that need structured AI model intelligence.

1. npm Packages

The repository publishes four reusable libraries:

  • @basemodel/schema for canonical schemas and TypeScript types.
  • @basemodel/registry for reading and writing canonical registry data.
  • @basemodel/intelligence for search, alternatives, and cost heuristics.
  • @basemodel/publisher for generating the public JSON datasets.

Install the packages you need:

npm install @basemodel/schema @basemodel/intelligence

Schemas and Types

import { ModelSchema, type Model } from '@basemodel/schema';

const rawData = { /* ... */ };
const parsed = ModelSchema.safeParse(rawData);

if (parsed.success) {
  const model: Model = parsed.data;
  console.log(model.name);
}

Intelligence Engine

IntelligenceEngine can load the registry from Node.js or hydrate from an already loaded snapshot in environments without filesystem access.

import { IntelligenceEngine, calculateCostEfficiency, searchModels } from '@basemodel/intelligence';

const engine = new IntelligenceEngine();
await engine.init();

const matches = searchModels(engine, {
  providerIds: ['openai'],
  modalities: ['image'],
  flags: ['vision_support'],
  minContextWindow: 100000,
});

const cost = calculateCostEfficiency(engine, 'openai/gpt-4o');
console.log(cost.tier);

In browser-like environments, hydrate the engine manually:

engine.hydrate({ models, providers, capabilities, pricing });

2. CLI

The CLI exposes the same intelligence logic from the terminal.

basemodel search --provider openai --modality image --flag vision_support --min-context 100000
basemodel info openai/gpt-4o
basemodel alternatives anthropic/claude-3-5-sonnet

Supported search filters are:

  • --provider
  • --modality
  • --flag
  • --min-context

3. Direct JSON Consumption

The publisher writes static datasets to dist/. Consumers can fetch those files directly from the repository or from any distribution mirror.

  • models.json - Models and their capabilities.
  • providers.json - Provider metadata.
  • capabilities.json - Canonical capability vocabulary.
  • licenses.json - License metadata.
  • apis.json - Model access methods.
  • benchmarks.json - Benchmark results.
  • pricing.json - Pricing records.
  • intelligence.json - Derived cost and alternative data.
import requests

url = 'https://raw.githubusercontent.com/ngodingsendiri/BaseModel/main/dist/intelligence.json'
response = requests.get(url)
data = response.json()

for record in data['intelligence']:
    print(f"{record['model_id']} - {record['cost_tier']}")

4. Gateway Plugins

Gateway plugins live in packages/collectors/src/gateways/. The collector validates plugin paths, runs plugins in isolated workers, and only injects the secrets registered for that gateway.

If you add a new plugin, update the secret registry and the security doc at the same time.