Deep Water CLI

The personal deep-research CLI. Pay only for what you use. $0.10 minimum per job. Full cost breakdown on every invoice.

npm install -g @unlikeotherai/deepwater
deepwater config --set api-key=dw_xxx
deepwater research "Compare the top 5 vector databases by query latency"

Who this is for

This CLI is the best way to use Deep Water as an individual. Ad-hoc research from your terminal, scripts, editor integrations, cron jobs, personal automation. One-off questions. Side projects. Technical due diligence before a purchase.

Running a team or production workload? Use the hosted API directly. It's the same engine, built for enterprise and heavy users: webhooks, job queueing, bring-your-own-keys, per-project billing, higher rate limits, and an SLA. See deepwater.live or docs.deepwater.live.

What it does

deepwater is the command-line client for the Deep Water research API. It runs multi-phase research jobs (scope → gather → synthesise → verify) against the hosted engine and streams back a cited markdown report.

• Four depth presets — light, standard, deep, heavy — each with its own URL budget and hard cost cap
• Usage-based billing — real cost plus a flat 50% margin, $0.10 floor. No tokens to buy, no packs to forget.
• Every claim cited — the report is grounded in the URLs the engine actually fetched
• Hard cost caps — jobs abort cleanly at the cap, you are never billed for work the engine did not deliver

Installation

npm install -g @unlikeotherai/deepwater
# or
pnpm add -g @unlikeotherai/deepwater

Requires Node.js 18.17 or later.

Configuration

Get an API key at deepwater.live and store it locally:

deepwater config --set api-key=dw_xxx

The config file lives at ~/.config/deepwater/config.json (mode 600).

You can also pass the key via environment variable:

export DEEPWATER_API_KEY=dw_xxx

Commands

deepwater research <query>

Run a research job and print the report.

deepwater research "Latest pricing tiers for Vercel Pro" --depth light
deepwater research "Full landscape of the vector DB market" --depth deep --output report.md

Options:

| Flag | Description | |------|-------------| | --depth <value> | light, standard, deep, or heavy (default: engine picks based on query) | | --format <markdown\|json> | Output format (default: markdown) | | --output <path> | Write to a file instead of stdout | | --report-model <name> | Override the final synthesis model | | --task-model <name> | Override the small task model | | --max-urls <n> | Override the acquisition ceiling | | --max-rounds <n> | Override the number of gather rounds | | --max-sub-questions <n> | Override the decomposition breadth | | --max-output-tokens <n> | Override the output ceiling | | --min-output-tokens <n> | Override the minimum report length | | --webhook-url <url> | Completion webhook | | --no-wait | Submit and print the job ID; do not poll |

deepwater status <job-id>

Fetch the current snapshot of an existing job.

deepwater config [--set key=value ...] [--json]

View or update local config. Supported keys:

• api-key
• base-url (defaults to https://api.deepwater.live)
• report-model
• task-model

deepwater version

Print the CLI version.

Environment variables

• DEEPWATER_API_KEY — overrides the configured API key
• DEEPWATER_BASE_URL — overrides the API base URL
• DEEPWATER_CONFIG — overrides the config file path

Docs

Full API reference and CLI guide: docs.deepwater.live

License

MIT. See LICENSE.