@openforge-ai/adapters

LLM provider adapters for Forge agent infrastructure.

Install

npm install @openforge-ai/adapters

Adapters

LLM Provider Adapters

These extend BaseLLMAdapter and handle model name validation and API key resolution. Their deploy() methods return the provider's base URL — they validate configuration, not manage infrastructure.

| Adapter | Provider | Model Validation | Env Var | |---|---|---|---| | AnthropicAdapter | anthropic | claude-* | ANTHROPIC_API_KEY | | OpenAIAdapter | openai | gpt-* or o[0-9]* | OPENAI_API_KEY | | GoogleAdapter | google | gemini-* | GOOGLE_API_KEY | | OllamaAdapter | ollama | any (local) | — |

Runtime Adapters

These implement RuntimeAdapter with full lifecycle management: deploy, status, and destroy.

DockerAdapter — Builds a Docker image from Forge config, runs it as a container with a health endpoint. Supports registry push, custom networks, and environment variables. Generates provider-specific runner scripts for Anthropic and OpenAI models.

AgentProviderAdapter — Generic REST adapter for hosted agent platforms. Communicates via a standard API pattern (POST /agents, GET /agents/:id, PUT /agents/:id, DELETE /agents/:id, POST /agents/:id/run). Works with any platform that follows this convention.

Usage

import { AnthropicAdapter, DockerAdapter, AgentProviderAdapter } from "@openforge-ai/adapters";
import type { ModelConfig } from "@openforge-ai/sdk";

const model: ModelConfig = { provider: "anthropic", name: "claude-sonnet-4-5-20251001" };

// Validate a model name against its provider
const adapter = new AnthropicAdapter();
adapter.validateModel(model); // true

// Deploy as a Docker container
const docker = new DockerAdapter({ runtime: "node:20-alpine" });
const result = await docker.deploy(model, { name: "my-agent", port: 3000 });

// Deploy to a hosted agent platform
const hosted = new AgentProviderAdapter({
  endpoint: "https://agents.example.com",
  apiKey: "...",
});
const agent = await hosted.deploy(model, { name: "my-agent", model });

Interfaces

• AdapterInterface — Base: validateModel() and deploy()
• RuntimeAdapter — Extends with destroy() and status()
• DeployResult / StatusResult / DestroyResult — Standardized return types

Links

• GitHub
• @openforge-ai/cli — CLI tool
• @openforge-ai/sdk — Types and SDK

License

MIT