v0.11.0

Agents

Securo now has a built-in AI layer — self-hosted, privacy-first, fully opt-in.

What's new

• Multi-provider LLM — OpenAI, Anthropic, Ollama, OpenAI-compatible (LM Studio, etc.). API keys encrypted at rest.
• MCP tool layer — agents read your accounts, transactions, categories, payees, budgets, reports, groups, splits, settlements; write actions go through a "proposal → approve" flow.
• RAG knowledge base — upload docs per agent, chunked + embedded (bundled fastembed or Ollama/OpenAI). Pin docs for always-in-context use. Powered by pgvector.
• Global chat panel — ⌘J from anywhere. Auto-injects the page you're on so the agent has context.
• Per-agent config — provider, model, temperature, prompt, history window, RAG top-N + similarity, tool whitelist, default flag.
• Streamlined nav — removed the Dashboard menu item; the Securo logo now links to the dashboard.

Opt-in setup

Add to .env:

AGENTS_ENABLED=true
COMPOSE_PROFILES=agents

Then docker compose up -d — both flags are needed (the first enables the app feature, the second tells Compose to start the mcp-server container). Off by default; zero cost when off.

Notes

• Migrations 046–049 add the agents tables. Migration 046 now refuses to run with a clear error message if pgvector isn't available, instead of crash-looping the backend.
• cryptography==46.0.7 pinned to fix SIGILL on Apple Silicon.

Full changelog: v0.10.4...v0.11.0

⚠️ Upgrading from v0.10.x

This release requires pgvector in the Postgres container. If your db service is on postgres:*, swap it:

  db:
-   image: postgres:16-alpine
+   image: pgvector/pgvector:pg16

Then docker compose pull db && docker compose up -d. Your pgdata volume is preserved (same pg16 major).