Run it on your own machine, free

Install a lightweight agent and workflows run on your hardware: no compute charge, data never leaves, reaching intranet and localhost the cloud can't. Or hand it to the managed cloud.

Workflows that run on your local or BYO agent are never billed for compute and consume no cloud time allotment; you provide the machine.

That holds on every plan, including Free: local and BYO runs are free and unlimited: never metered, never billed, never counted against any quota. Paid plans buy cloud capacity, not local permission.

Local and BYO runs cost $0 in compute on every plan. The cloud is the metered, managed convenience layer. · Every plan, including Free: unlimited local and BYO runs. Paid plans add included cloud running time, faster monitor cadence, and bigger fleets. See the pricing page for what each plan includes.

Local agents are unlimited on every plan; you bring the machines. Cloud agents are plan-gated. On Free and Starter your one "cloud agent" is your own device acting as the cloud agent; a genuine multi-agent fleet begins at Growth.

Local and BYO agents are never counted against this. See the pricing page for the full fleet limits.

Because it runs on your machine, a BYO agent reaches systems no cloud browser can: internal and intranet sites, localhost apps, LAN devices, VPN-only and on-prem systems.

The agent runs where your systems live. Build a workflow that automates an internal-only web app, then call it over REST or expose it as an MCP tool while the agent stays inside your network.

Configure your own AI provider keys (Anthropic, OpenAI, Ollama, or any OpenAI-compatible endpoint) directly in the agent.

Those keys live in the agent's local config on your own machine; the agent calls the provider directly, so your keys are never sent to or stored on Writ's servers. AI that runs on your own keys through your own agent is unbilled by Writ: you pay your provider directly, or run Ollama locally for zero per-token cost.

# Lives on YOUR machine. Never sent to Writ.
ai:
  provider: anthropic        # anthropic | openai | ollama | openai-compatible
  api_key: "sk-ant-•••••••••" # your key, your bill
  model: claude-opus-4

# Or run entirely on local hardware:
# ai:
#   provider: ollama
#   base_url: http://localhost:11434
#   model: llama3.1

Point the agent at a local Ollama model and your workflow's AI steps run entirely on your hardware: no data leaves your machine, no per-token cost. Use it for fully private automation where the page content never touches a third party.

When you'd rather not run hardware (always-on schedules, runs while your laptop is closed, scale across many sites), the managed cloud fleet runs the same workflows with nothing to install.

Cloud runs are metered by running time from your plan’s included usage, at roughly $0.12/browser-hour standard. Residential bandwidth comes out of that same pool, at 7 credits per megabyte ($7/GB) — there is no separate egress allowance to forecast. The cloud sells convenience and scale, not features. Every workflow runs the same on local or cloud.

Add funds to your wallet from $5 - $1,000. See pricing.

For each workflow you choose where it runs: your local agent, the shared cloud pool, or a gVisor-isolated cloud venue for sensitive runs that touch credentials or secrets.

Isolation is a choice, not an afterthought. How isolation works.

Agents advertise a speed class (fast, throughput, or balanced) and a performance score. Writ routes latency-sensitive work to fast machines and bulk runs to throughput machines, automatically.

Local and BYO agents are free of compute on every plan; you bring the hardware. The cloud is the metered, managed convenience layer on top.

Local and BYO runs are free and unlimited on every plan, including Free: never metered, never billed. Cloud runs are metered by running time. · Want to run the whole stack yourself? See self-hosting.

Writ runs on your own accounts, with your own credentials and data, on sites you are authorized to use.

Trust centerSecurity