Provider Access

This page explains what a user actually needs to run MindStore and how provider authentication is expected to evolve.

What A User Needs Today

Required

• PostgreSQL 16+ with pgvector and pg_trgm

Optional

• Vercel or any other host
• Google OAuth
• API keys for hosted AI providers

Important Clarification

MindStore does not require Supabase specifically.

Supabase is one possible way to host PostgreSQL, but it is not the product requirement. A user can run MindStore with:

• self-hosted PostgreSQL
• Supabase
• Neon
• Railway
• Render
• a VPS with Postgres installed

MindStore also does not require Vercel specifically. Vercel is just one deployment option.

What Works Without AI Credentials

Today, a user can still:

• import data
• browse memories
• use docs and MCP setup surfaces
• use keyword and metadata-driven exploration

without a paid hosted AI provider.

What Needs AI Access

These features are strongest when a provider is configured:

• semantic embeddings
• RAG chat
• some analysis and generation plugins
• higher-quality cross-memory reasoning

Supported Access Modes Today

• API keys for hosted providers such as Gemini, OpenAI, OpenRouter, and custom OpenAI-compatible endpoints
• Local runtime access through Ollama

Planned Access Modes

MindStore should grow beyond API-key-only flows.

The planned direction is:

1. Keep API keys and local runtimes first-class because they are stable and easy to self-host.
2. Add encrypted auth profiles for providers that expose supported OAuth or device-login style flows.
3. Let provider adapters and plugins register their own auth flows instead of hardcoding all auth logic in core.

Why Subscription Auth Matters

Many users already pay for AI plans and do not want to manage separate API billing just to use personal-memory features.

That means MindStore should support, where technically and contractually safe:

• account-based provider auth
• device code login
• OAuth-backed token exchange
• trusted local bridge adapters

Current Recommendation

For now:

• Use Gemini if you want the simplest low-cost hosted option.
• Use Ollama if you want a local/private stack.
• Use OpenAI or OpenRouter if you want broader model choice and do not mind API billing.

Next Technical Steps

• Standardize provider capability metadata in the app and docs
• Add encrypted auth-profile storage
• Add provider auth adapters for supported subscription-style flows
• Keep MCP and plugin runtime provider-aware so auth modes can evolve without rewriting the whole product