AI PLATE/ Docs
Ctrl K
Back to AI Plate
Documentation / AI providers
YOUR AI PLATE HANDBOOK

Chat, embeddings & local providers

Configure the two model paths independently, including Mistral and local servers.

Desktop source reviewed October 5, 2026
On this pageChat and embeddings are separateBuilt-in endpoint configurationVerify a local setupCheck the right failure point

Chat and embeddings are separate

SettingUsed forCheck it with
Chat provider/modelConversation and tool-call generationA short message, followed by a small enabled-tool task.
Embedding provider/modelDocument indexing and similarity retrievalA small document, then a question about a specific passage.

Choose both in Models & Reasoning. A successful chat response does not demonstrate embedding support. Available model names must match your provider account or the models loaded on your local server.

Built-in endpoint configuration

Provider IDDefault endpoint in this checkoutConfiguration
mistralhttps://api.mistral.ai/v1MISTRAL_API_KEY; optional MISTRAL_BASE_URL override.
ollamahttp://localhost:11434/v1Start the local server; optional OLLAMA_BASE_URL override.
lmstudiohttp://localhost:1234/v1Start its local server and load a model; optional LMSTUDIO_BASE_URL override.

These are the defaults implemented by AI Plate’s registry, not a statement about current external model availability. Mistral uses the OpenAI-compatible adapter; Gemini and Anthropic have their own adapters.

Verify a local setup

  1. Start the local model server and load or install the intended chat model.
  2. Select its provider and exact model name in AI Plate.
  3. Send a short greeting, then verify a tool call with a model that supports tools.
  4. Configure a local embedding model separately and test one document.
  5. Keep both model paths local if you intend to avoid cloud inference. Downloads, web tools, and external connectors may still use the network.

Check the right failure point

FailureLikely check
Connection refusedLocal server running, host, port, and base URL.
Model not foundExact provider model ID or the model loaded by the local server.
Chat succeeds; tools failTool-calling support, enabled tools, and chat mode.
Document ingestion failsEmbedding endpoint, model, and credentials.
Implementation sources

Paths in your AI Plate source checkout:

  • providers/provider-registry.ts
  • providers/provider-factory.ts
  • core/embedder.ts
  • core/config.ts
  • ui/index.html
← Previous guidePersonal memory & dossierNext guide →Local development
AI Plate product & developer documentation · Static reference, no API keys collected.