A Glean alternative that runs in your basement, not Big Tech’s cloud

What it does

Xyne is a self-hosted search and answer engine that ingests data from workplace SaaS apps—currently Google Workspace including Gmail, Drive, Calendar, and Contacts—and lets employees query it with natural language. It returns answers with source citations and enforces the original permission models from each connected app, so users only see results they already have access to. The project aims to expand into an “Org Operating System” that feeds context to both humans and AI agents.

The interesting bit

Most enterprise search tools treat “Google integration” as Drive-only; Xyne actually ingests email, attachments, and calendar events. It is also aggressively model-agnostic, letting you point it at OpenAI, Anthropic, AWS Bedrock, or a local Ollama instance without rewriting queries.

Key highlights

• Self-hosted by default: runs locally, on-prem, or on a cloud VM via Docker.
• Permissions-aware search: live enforcement of existing app-level access controls.
• Model agnostic: plugs into commercial APIs or local models like DeepSeek via Ollama.
• No telemetry or training on customer data, according to the project’s claims.
• Multi-threaded ingestion pipeline for indexing connected sources.

Caveats

• Integrations beyond Google Workspace are still marked as “NEXT” on the roadmap; Slack and Atlassian support is not yet available.
• At 671 GitHub stars, the project is early-stage and enterprise users should expect rough edges around scaling and connector maturity.

Verdict

Teams that want Glean-like AI search but refuse SaaS black boxes should kick the tires. If you need mature connectors for Slack, Jira, or GitHub today, wait a quarter.