Skip to main content
The CLI runs the nine research jobs, reaches the full tool catalog, sets up your clients, signs you in, and serves MCP over stdio. Run hive --help for the list on your installed version, or hive <group> --help for one group.

Job commands

Each of the nine tools has its own command. All of them accept --json. The other parameters are flags, named after the tool’s parameters in kebab-case or snake_case, with the same rules as hive tools call below. Arrays take a comma-separated list or a JSON array:
Every job command needs a signed-in account: run hive auth login, or set HIVE_API_KEY.

Setup

--all also installs the hive CLI globally (npm install -g hive-intelligence, the same version as the CLI you ran) and the hive and hive-setup skills. Pass --skip-global to leave the CLI off your PATH, and --skip-skills to configure the clients without the skills. If the global install is not permitted, init says so and carries on; the skills then run Hive through npx. Skills are installed with the shared skills installer (npx -y skills@1.7.6 add hive-intel/hive-sdk --global --yes); if that fails, init copies them itself. Pass --skip-verify to keep the whole command non-interactive, which is what an agent should run. Both init and uninstall accept --dry-run if you want to see what would change first. hive init never reads or writes AGENTS.md or CLAUDE.md. The routing block is written only by hive routing install, and HIVE_ROUTING=never disables that on a machine.

Tools

The full catalog, about 420 tools beyond the nine, lives under hive tools. Each catalog call costs 1 credit.
hive tools search matches literal text, not intent. Search for a keyword like funding or nft, not a phrase like “what is this token worth”. A phrase that does not appear verbatim in a tool’s name or description returns no results.
hive tools list returns 50 tools at a time. Use --limit and --offset to page through the full catalog.

Authentication

Only hive auth login takes --profile <name>. The other four commands act on the profile that is already active, and reject --profile with error: unknown option. Use hive auth switch <name> to change which one that is. There is no signup command. hive auth login opens the browser, and an account is created on first sign-in if you do not already have one.

Keys

Key management lives in the dashboard, at hiveintelligence.xyz/dashboard/keys. Most clients never need a key at all. Claude, Claude Code, Cursor and VS Code authorize through the browser on first connect, and hive auth login stores a credential for the CLI. A key is for the case with no browser to click: a server, a script, or CI.

Other

serve is for clients that only speak stdio, or for local development against your own provider keys. Most users should connect to the hosted endpoint instead.

Removed in 1.19.0

Scripting

--json gives output you can pipe:
The envelope has ok, data, and meta, where meta carries the tool name, provider, and runtime status.
--json is honored by the nine job commands, by hive tools list, search, info, and call, and by hive doctor and hive status. Commands that only print human-readable text, including everything under auth, ignore it and write to stderr. Piping those to jq gives you an empty stdout, not an error.
Useful global flags: --jq is a small built-in evaluator, not a bundled copy of jq. It supports .field, .a.b, .[N], .[], and select(.field op value). For anything more, pipe --json into real jq. -q suppresses everything that is not data, including the tools table. hive tools list -q prints nothing at all unless you also pass --json.

Exit codes

Exit codes are stable, so hive doctor works as a health check in CI: