# Laya

> Laya is an open-source, local-first AI notification command center that aggregates Slack, Gmail, GitHub, Jira, Notion, Outlook, and Calendar notifications — powered by local LLMs via Ollama and LM Studio, or cloud models like Claude and GPT with your own API keys. Notifications are processed through a multi-stage AI pipeline and delivered as ready-to-approve Action Cards.

Laya connects to platforms like Gmail, Slack, Jira, GitHub, Bitbucket, Google Calendar, Outlook, Linear, and Notion. Events are ingested via n8n workflows, classified by AI personas (Engineer, Comms, Ops, Sales, HR, Finance), and synthesized into intelligence reports with pre-drafted actions — all before the user opens the app.

## Architecture

- Desktop app: Tauri (Rust) + SvelteKit (Svelte 5) frontend
- Engine: Python FastAPI backend with async SQLite + ChromaDB
- Integration gateway: n8n running as a local Node.js process
- All data stays on the user's machine (local-first)
- LLM calls go through LiteLLM; supports Anthropic, OpenAI, Google, OpenRouter, Ollama, LM Studio, any OpenAI-compatible endpoint, and installed CLI coding agents driven as inference backends

## Pipeline

Events flow through: Ingest → Space Resolution → Rules → Router (fast LLM classification) → Persona Workers → Stager (strong LLM synthesis) → Emit → Post-emit steps (entity resolution, context grouping, ChromaDB indexing, group summaries, Omni updates).

## Key Features

- **Action Cards**: AI-generated cards with intelligence reports, priority classification, and one-click actions
- **6 AI Personas**: Engineer, Comms, Ops, Sales, HR, Finance — each with specialized processing
- **Coherence**: Cross-platform entity search using hybrid retrieval — dense vectors (ChromaDB) + lexical BM25 (SQLite FTS5) + entity search merged via Reciprocal Rank Fusion, with AI-generated narratives
- **Omni**: Rolling cross-platform summary with four temporal layers (Attention, Recent, Period, Milestone) and progressive summarization
- **Context Association**: Automatic semantic grouping of related cards across platforms; learns editable natural-language context rules from user corrections (auto-consolidated by an LLM)
- **Coding Agents**: Run Claude Code, Gemini CLI, Codex, Pi CLI, or Cursor Agent directly from any card with a dedicated workspace
- **Agent inference backends**: Run any pipeline stage on an installed CLI agent's own quota via `agent/<id>/<model>` model ids (native schema enforcement on Claude Code; best-effort + retry on others)
- **Agent Studio**: Full agent configuration — runtime, execution mode, working directories, detailed prompts
- **Composer**: Cross-platform outbound actions from a single interface (9 platform backends, plus a direct SMTP backend)
- **Daily Briefings**: Synthesized morning briefing with events, action items, and key updates
- **Spaces**: User-defined contexts with per-space model configs, API keys, and event routing
- **Chat**: Hybrid (vector + BM25) search across all ingested events with RAG-powered retrieval; can also create/edit filter, classification, and processing rules
- **Smart Classification & Rules**: Multi-model pipeline that learns from corrections; user-authored filter and processing rules with optional AI-evaluated conditions and a searchable firing log
- **Budget Tracking**: LLM cost monitoring by feature and pipeline step with monthly caps; plus window-based usage budgets for agent inference backends
- **Self-Hosted & On-Prem**: Cloud and self-hosted instances (Bitbucket Server / Data Center, GitHub Enterprise) via a per-repo host field
- **Privacy Tiers**: Three-tier system with optional local-only mode for sensitive data

## Integrations

Gmail, Slack, Jira, GitHub, Bitbucket, Google Calendar, Outlook, Linear, Notion. Custom integrations via n8n workflows posting to the Laya Event format.

## Links

- Source code: https://github.com/aayushch/laya
- Documentation: https://aayushch.github.io/laya/docs.html
- Privacy policy: https://aayushch.github.io/laya/privacy.html
- Terms of service: https://aayushch.github.io/laya/terms.html