Copy this into any coding agent. It fetches the runbook and does the rest.
Launch it from a link
These agents publish a URL scheme that carries a prompt, so one click opens them with the line above already typed. Nothing is auto-sent - you still press Enter.Deep links need the agent installed locally with its URL handler registered. If a link does not open your client, copy the prompt into it instead.
What the agent actually does
Working throughvalyu.ai/agents.md, the agent:
1
Installs the CLI
curl -fsSL https://get.valyu.ai | bash on macOS/Linux, or Homebrew, npm, or PowerShell. It checks for an existing install first rather than blindly reinstalling.2
Runs the login flow
valyu login opens a browser device flow that mints a scoped key - there is nothing for you to copy and paste. With --json the CLI streams device_code, auth_waiting, and auth_success events, so the agent can show you the user code and tell you it is waiting on you.3
Verifies before claiming success
valyu doctor and valyu account balance, then a single query against a source that works on every plan. The runbook tells the agent not to report success without this, and not to run paid DeepResearch tasks or top up your balance on its own.4
Installs the skill
npx skills add @valyu/cli teaches the agent the full command surface, so it stops guessing at flags. See CLI agent integration.5
Runs a real research task
After setup, ask the agent to run a research task you choose. Specify the scope and output you want, review the mode and expected charges, then inspect the resulting report and sources.
Build Valyu into your app
To have your coding agent write Valyu into your own codebase, start from the playground. Set up the request you want, then use Build with your agent above the settings: one click copies a prompt that carries your exact request and points the agent at platform.valyu.ai/integrate.md, and its menu opens Claude Code, Codex or Cursor with the prompt already filled in. The agent then:- signs you in with one browser approval and writes a new key, capped at 10 USD a month and named after your project, straight into your env file, so no key is pasted or printed
- picks the client for your stack:
valyu-js, the Python SDK, the Vercel AI SDK tools, LangChain or plain HTTP - writes a small server-side module, tests with the API mocked and one live check, then reports the files it changed and how to set the key in production
Agent with its own inbox?
If your agent has an AgentMail inbox, it doesn’t need your account at all: it can open its own with Continue with AgentID, get its own key and free credits, and share them with your other agents. See Sign in with AgentID.Prefer to do it yourself
CLI
Install, authenticate, and run search, content extraction, and deep research from your terminal.
Remote MCP
Hosted server with six tools. Drops into Claude, ChatGPT, Cursor, or any MCP client.
Quickstart
First API request in minutes with Python, TypeScript, or REST.