Your whole Glean workspace, from the terminal.
Chat, search, agents, tools, insights, and near-complete Indexing API coverage — in a fast local REPL inspired by Claude Code. Pure Python. Zero runtime dependencies. Fully usable before you even log in.
| ⚡ No install ceremony | python3 install.py builds an ~84 KB single-file zipapp, puts glean on your PATH, and — on macOS — makes it Spotlight-launchable. Still zero dependencies |
| 🧪 Offline by default | Seventy interlinked documents across five faux datasources back every mock response, so the whole CLI is explorable with no tenant, no token, no network |
| 🔑 Browser SSO | /login acme runs OAuth 2.1 + PKCE through your normal SSO. Or bring a token — secure refs keep the real secret in an env var, never on disk |
| 🗣️ Talk to it in English | ?search for the Q2 plan and summarise it — Glean Assistant writes the commands, the CLI validates them, and anything destructive waits behind one confirm |
| 🗺️ It remembers what you investigated | /flow links today's incident to last month's renewal because a cited doc mentioned it in passing — with the evidence attached |
| 🏠 Your own files, indexed locally | /personal index ~/Documents builds a portable SQLite index — .md, .html, .json, even .docx/.xlsx/.pptx — then /mode local searches it. One file, no server, no network |
python3 -m glean_code # straight from the repo
# or
python3 install.py && glean # installed, with a macOS Spotlight app/login acme
/search "quarterly planning"
/chat "summarise the Q2 plan"
/login acme opens browser SSO; /login --token <bearer_token> takes a Glean-issued token instead. No login at all? You're still up and running — mock mode serves ranked results from a real corpus.
→ Install guide · Configuration · Command index
| Command | What it does |
|---|---|
?login into acme and search for "Q2 plan" |
Natural language, planned and confirmed before anything runs |
/search "checkout incident" --datasource jira |
Search every connected source |
/chat "what changed in the pricing doc?" |
Threaded Glean Assistant chat |
/agents.run <agent-id> "draft the release notes" |
Run agents, call tools |
/insights --all --export insights.csv |
Usage metrics straight to CSV |
/debug.user gdrive alice@example.com |
Why can't Alice see that doc? |
/index.bulk-documents --path ./docs/ --datasource custom1 --object-type Article --dry-run |
Index a folder, inspecting the payload first |
/metadata.attach --doc-id <id> --group tickets --values owner=alice |
Enrich docs without re-uploading them |
/personal index ~/Documents --label docs |
Index your own files, locally |
/graph "quarterly planning" --html graph.html |
Graph how that content relates — documents, authors, sources |
/flow show |
Draw the investigations you ran |
|
Search, autocomplete, recommendations, threaded chat, summarize, answers, feedback — plus docs, people, entities, announcements, collections, pins, Go Links, and verification.
|
List and run agents, list and call tools with inline JSON arguments, and read the results back in the terminal. |
|
Read/debug toolkit, single-record writes, bulk + paged uploads, process-all rebuilds, custom metadata, and |
MAU/WAU, sign-ups, search satisfaction, clicks by datasource, Assistant and Agent activity — |
|
|
|
Every command takes the same path: parsed, dispatched to a handler, turned into one REST call. The mode decides who answers it — and mock and local replies are labelled in the output, so you always know which index you are reading.
/flow records the investigations you run — every /chat and /search, and the documents they cited — then finds the connections between them. Not only "both mentioned INC-1183", but the indirect case: two conversations sharing no vocabulary, linked because a document cited by one refers to the other's subject in passing.
Sessions run down a rail in the order you worked; each connection branches off it carrying the evidence that earned it, so every link can be read rather than taken on trust. /flow timeline renders the same graph as a self-contained HTML page. Local SQLite, mock-only by default — recording live data is opt-in.
Mock mode proves every command works offline against a fictional corpus. /personal points the same machinery at content that is actually yours.
| Command | What it does |
|---|---|
/personal index ~/Documents --label docs |
Build the index from a folder |
/personal search "salary bands" --explain |
Which terms hit, which missed, and the bm25 score |
/personal link && /personal related roadmap |
A phrase graph, with shared phrases as evidence |
/mode local |
/search and /chat now answer from your files |
SQLite FTS5, incremental on a content hash, .docx/.xlsx/.pptx read straight out of their ZIP-XML with the stdlib. No server, no daemon, no Docker, no network, no credentials — the whole index is one file you can copy between machines. Answers are labelled [LOCAL INDEX] and quote your passages verbatim; the REPL has no model in-process and will not invent prose. Four MCP tools expose the same index to an agent that does.
→ How-to guide · Reference & design notes
Released. The full Glean Code REPL — slash commands, status bar, mock/live switching, secure-token storage — in the editor sidebar of VS Code, Cursor and VSCodium. Run searches, kick off agents, and pin docs without leaving your code window.
One line installs the latest release into every VS Code, Cursor and VSCodium it finds — no clone, no Node, no build: install instructions.
/graph renders as an interactive tab — click a node for its edges, drag to move, scroll to zoom:
The same extension in VSCodium:
| 🚀 Install | Run from the repo, the zipapp installer, the macOS app, first run |
| ⚙️ Configuration | Auth methods, every config key, modes, files on disk |
| 🗂️ Command index | Every command, grouped by surface |
| 📖 Command reference | Usage, parameters, examples, endpoints |
| 💬 Natural language | How /ask plans, validates, and confirms |
| 🧱 Indexing · Metadata | Debug toolkit, writes, bulk, process-all, --path, custom metadata |
| 📊 Insights | Flags, output, and CSV export |
| 🧪 Mock corpus | The offline corpus — inventory, ranking, bring-your-own format |
| 🏠 Local indexing · Personal | Index your own folders, search them, keep them current |
| 🕸️ Knowledge graph | How /graph builds nodes and edges, and how shared terms are scored |
| 🗺️ Flow mapper | Capturing investigations, linking them, retention questions |
| 🔐 SSO / OAuth · Secure tokens | Browser sign-in, secure refs, the masking matrix |
| 🔌 MCP server | Glean as native tools in Claude Code, Claude Desktop, Cursor |
| 🏛️ Architecture · REST paths | Module map, request flow, endpoints, how to add a command |
| ✅ Testing | Running the 1,151-test suite and what it covers |
| 🛟 Support · Changelog | How to report a bug · release history |
Note
ℹ️ Known API incompatibilities. Twelve Client API commands currently diverge from Glean's published OpenAPI spec — seven return 404/405/400 against a live tenant. The Indexing and Custom Metadata surfaces are unaffected, and mock mode masks all of it. Full detail, locations, and suggested fixes: incompat_report.md.
MIT © 2026 barkz