Chat
Interactive sessions grounded in your real environment: installed components, health score, workspace, git status.
AI Assistant
Most AI tools guess at your environment. DevForgeKit's assistant reads it: every prompt is grounded in real installed components, compatibility state, and workspace context.
Capabilities
Interactive sessions grounded in your real environment: installed components, health score, workspace, git status.
ai doctor turns a symptom into a structured diagnosis: summary, reason, fix, estimated time, and risk.
Explain any tool or error; review configs and setups with environment-aware context.
ai generate scaffolds through the same audited project generator as devforgekit new, never freehand files.
ai planner maps a natural-language goal onto real registry collections, recipes, and packages; invented names are filtered out, never acted on.
ai repair executes through the same confirmation-gated repair engine as compatibility repair.
In action
devforgekit ai: zsh
ReadyProviders
Cloud when you want scale, local when you want privacy. Providers with the same wire format share one client implementation.
| Provider | Type | API key | Notes |
|---|---|---|---|
| OpenAI | Cloud | OPENAI_API_KEY | Chat, streaming, and embeddings; powers the registry's semantic search index. |
| Anthropic | Cloud | ANTHROPIC_API_KEY | Native integration with a separate system field and typed SSE streaming events. |
| Google Gemini | Cloud | GEMINI_API_KEY | Chat, streaming, and embeddings support. |
| Groq | Cloud | GROQ_API_KEY | Very fast inference over the OpenAI-compatible wire format. |
| OpenRouter | Cloud | OPENROUTER_API_KEY | One key, many models, routed through a single API. |
| Ollama | Local | none required | Fully local models. No API key, no data leaves your machine. |
| LM Studio | Local | none required | Local model server with a desktop UI. |
Configuration
Configure a provider
# ~/.config/devforgekit/config.yaml
aiProvider: anthropic
aiModel: claude-sonnet-5
# aiEndpoint: http://localhost:11434 # for Ollama / LM StudioPrivacy
Conversation turns live in memory only and are discarded when a session ends. The local event log records what happened (command, provider, duration), never what was said.
API keys come from environment variables or your workspace's declared secret reference, never stored in plaintext config, never sent anywhere except the provider you chose.
Ollama and LM Studio run models entirely on your machine. No API key, no network, no data leaving localhost.
Configure any of seven providers in two commands, or run fully local with Ollama.