LoginSign Up
← Back to all sparks
B

Browser Use

DEVOPS
Velocity2.5

Browser Use hardens its cloud agent API while its own model tracks frontier accuracy.

browser-automationai-agentsdeveloper-apiself-hosted-modelsecuritypricing
◆Current state
Browser Use is an AI browser-automation platform that ships both a hosted agent API and its own task model (BU 2.0). Recent work splits between developer-control primitives — opt-in session recording, OAuth browser access, bring-your-own-key, sensitive-data handling — and the underlying agent and model themselves. The changelog reads like an infrastructure product maturing its API surface rather than chasing new demos.
◆Where it's heading
The arc runs from raw capability (the BU 2.0 model, the from-scratch Agent API) toward production controls: recording made opt-in, secrets kept out of logs, BYOK, and a free tier to widen the top of the funnel. Code Mode's shift from one-shot tasks to reusable scripts points the same direction — durable, repeatable automation instead of throwaway runs.
◆Prediction
Expect continued hardening of the v4 harness — more granular session controls and enterprise-grade data handling — alongside iteration on the BU model line to hold its accuracy-per-dollar edge against frontier APIs.

◆Recent moves

  1. 23d ago

    Recording Controls & OAuth Browser Access

    The v4 harness makes session recording opt-in rather than always-on and adds OAuth browser access for terminal flows. These are control-surface refinements — the API ergonomics that matter to teams running agents in production, not headline capability.

    View source ↗
  2. 4mo ago

    BYOK, Code Mode & Sensitive Data

    Three production-oriented additions land together: bring-your-own LLM keys, secrets handling kept out of logs and traces, and Code Mode, which returns a reusable Python script instead of a one-off result. Together they push the platform toward repeatable, auditable automation.

    View source ↗
  3. 4mo ago

    Free Tier, Agent Signup & New Pricing

    ⚡ SPARK

    Making Cloud free and letting agents self-provision accounts via CLI is a distribution move, not a feature — it lowers the barrier to embedding Browser Use inside other agent stacks, consistent with the platform's shift toward being infrastructure.

    View source ↗
  4. 4mo ago

    CLI 2.0 + Weekly Update

    CLI 2.0 brings Browser Use into the terminal alongside Claude Code and Cursor, syncing real browser profiles, while stealth work (Chromium 146, captcha solvers) chases benchmark leadership. Solid execution on the existing agent-in-your-workflow thesis.

    View source ↗
  5. 5mo ago

    BU Agent API & SDK 3.0

    ⚡ SPARK

    The experimental ground-up agent — pitched as 'Claude Code for the browser' — reframes the product from a scripting library into an autonomous multi-step worker, the clearest directional shift in the recent arc.

    View source ↗
  6. 6mo ago

    Browser Use Model - BU 2.0

    ⚡ SPARK

    Shipping their own model that matches Claude Opus 4.5 accuracy at 40% faster speed is the strongest statement that Browser Use wants to own the agent stack end-to-end, not resell frontier APIs.

    View source ↗