AI Providers and Your Data

Cloud AI is opt-in. Nothing is sent to any AI provider until you configure API keys and trigger an AI feature. Here is exactly what goes where.

What is never sent to AI providers

  • Your project files (unless you explicitly attach them as context)
  • Your semantic memory database and embeddings
  • Your search queries (search runs locally)
  • Your browsing history or browser extension activity
  • Your account information or license details
  • Content from projects you have not attached to the current task

What is sent when you use AI features

When you trigger an AI action — a chat message, an agent step, or any feature that calls a language model — the following is sent to your configured provider (not to ThinkForge):

  • Your prompt — the message you typed or the instructions the agent assembled
  • Attached context — any documents, memory chunks, or files you explicitly added as context for that request
  • System instructions — template text that guides how the model should respond
  • Recent conversation history — in chat mode, recent messages for continuity
  • Model settings — temperature, max tokens, and other generation parameters
  • Your API key — in the authorization header to authenticate with the provider

What agents send

When an agent runs, AI steps may include:

  • The input documents you attached to the deployment
  • Agent instructions and rules
  • Output from previous steps in the pipeline
  • Reference files you explicitly added as directives

Agents only send content that is part of their configured context. They do not send your entire project folder or memory database.

What stays completely local

  • Embedding generation — uses a local model on your machine (no API call)
  • Semantic search — runs against your local database
  • File indexing and chunking — local processing only
  • Desktop-extension communication — localhost HTTP only
  • MCP server — runs on localhost for local AI editors

ThinkForge never sees your AI traffic

Direct to provider. AI requests go directly from your machine to the provider (OpenAI, Anthropic, Google, or OpenRouter). ThinkForge does not proxy, intercept, log, or store these requests. We never see your prompts or responses.

You control what leaves your machine

Three factors determine what gets sent to an AI provider:

  1. Whether you have API keys configured — no keys means no AI calls, period
  2. Which features you use — only features that explicitly need a language model make API calls
  3. What context you attach — you choose which documents and files to include in prompts and agent inputs

If you want maximum privacy, you can use ThinkForge without any cloud AI keys. Local search, file management, organization, browser capture, and the Chrome extension all work without cloud AI. Add keys only when you want AI-powered features like chat, summarization, or agent execution.

Provider data policies

How each provider handles your data after receiving it is governed by that provider's terms and privacy policy — not by ThinkForge. When choosing a provider, consider their policies on data retention, training, and logging.

Related