Recipes for building on Glean
Runnable patterns that go from problem to a working demo to scaffolded starter code — with the architecture, auth, and permissions laid out for each.
On-call copilot with a real approval gate
Triage an incident from your own runbooks and past incidents, propose one pre-registered action, and let a human approve it — where the gate refuses the wrong person, expiry escalates instead of auto-approving, and every attempt is audited.
Open the build guideDiscover a published Glean agent's A2A card and run it from any A2A client — multi-turn, streamed, permission-aware.
The end-to-end showcase — index a developer catalog into Glean, then embed permission-aware search and chat back into the portal your team already uses.
The hello-world of Glean apps: one page, one input, one permission-aware cited answer — built two ways, with the Web SDK and with the Chat API.
Point Claude Code, Cursor, and Claude Desktop at your Glean MCP endpoint and run one enterprise task from each — same context, three surfaces.
One page per account — status, risks, and a drill-in chat — assembled from whatever your instance already knows about that customer. No CRM export, no separate index.
Put permission-aware Glean search and chat directly inside an internal app with the Web SDK, so your team gets answers where they already work.
Build a Glean agent that plans, retrieves, and acts through a governed custom tool — with a safe fallback when the tool is denied.
Prompt Lovable into a private, single-user IT helpdesk prototype on the Glean Chat API — zero hand-written backend.
Prompt Replit Agent into a private, single-user HR lookup prototype on the Glean Chat API — zero hand-written backend.
A guided first-week hub for new hires: a checklist with progress, and every step able to answer itself from your own onboarding content.
Use Glean's Platform API as the retrieval layer for your own LLM app — every result ACL-filtered for the caller before it ever reaches the model.
Turn a customer questionnaire into grounded, cited draft answers — where every claim carries a source, unsupported rows route to a human, and nothing reaches the customer without approval.