Skip to main content

​
Prerequisites

SQLite is used only for the automated test suite (the vector(N) columns swap to JSON text via the test migrations). Production and development run on PostgreSQL + pgvector.

​
1. Install dependencies

​
2. Provision PostgreSQL + pgvector

pgvector must exist in the target database before migrating:
Then point .env at it:

​
3. Choose AI providers

Chat and embeddings are configured separately — Anthropic, for example, has no embeddings endpoint, so you pair it with an embeddings-capable provider.
When AI_PROVIDER=anthropic and AI_EMBEDDINGS_PROVIDER is empty, AskMyDocs auto-selects the first embeddings-capable provider with a configured key in a dimension-safety-first order (openai → openrouter → regolo → gemini), so a deployment never silently corrupts ingest writes with a mismatched dimension. See AI providers for the full matrix.

​
4. Migrate

​
  1. Embedding dimensions

The embedding dimension is part of the schema contract. The stock vector(N) columns are sized for 1536 dimensions (openai/text-embedding-3-small, the default, which also matches OpenRouter’s default embeddings model). Choosing a different-dimension model — e.g. qwen/qwen3-embedding-4b at 2560 dims — requires all three of:
  1. migrating the vector(N) columns to the new dimension,
  2. flushing embedding_cache,
  3. re-indexing the corpus (re-ingest).
Mixing dimensions silently corrupts retrieval. Pick the embeddings model before your first ingest.

​
6. Run it

Open /app for the SPA (chat + admin), or hit the JSON API directly. Create your first admin user with the role seeder / auth:grant command (see Admin panel).

​
Scheduler

Several rotations run on the Laravel scheduler (bootstrap/app.php). Enable the system cron entry so they fire:
All commands accept --days=N and --project= overrides; 0 disables the corresponding rotation.

​
Next steps

Quickstart

Ingest and ask your first question.

Configuration reference

Every env knob, grouped by subsystem.