fastbrowse
Browser agent where Jev picks each action from indexed page controls, an LLM plans and reads, and answers cite page quotes.
Overview
fastbrowse gives an agent a browser, locally or in a cloud browser. It indexes the page into candidate controls and has Jev pick one, so it cannot click something that was never on the page, while an LLM plans and reads. Factual claims in a complete answer cite verbatim quotes from the page, and it can return schema-validated data. Code controls credentials, authorisation and spending, with scoped credentials and an authorisation gate.
The README shows a recorded run of 13 steps in 20.9 seconds costing $0.0034 in model calls and links a benchmark page comparing task success, speed and cost against Browser Use agents, noting that rows are comparable only at matching task versions and that navigation results do not establish a winner for every workflow. It runs via uvx with OpenRouter for Jev and the LLM, defaults to a Browser Use Cloud browser that passes bot checks, and can drive a local Chrome over CDP or Electron apps. A JavaScript interface is available.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which indexed page control should the agent act on next? | — |
How it works
- Index the page into candidate controls
- Jev picks the next action; an LLM plans and reads
- Execute the action in the browser
- Return a cited answer or schema-validated data
Strengths
- No invented selectors
- Verifiable citations for claims
- Credential and spending controls in code
Caveats
- Default cloud browser requires a Browser Use key
- Benchmark comparisons limited to matching task versions
Getting started
export OPENROUTER_API_KEY=..., then uvx fastbrowse "What is the title of the top story right now?" --start https://news.ycombinator.com/.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- CLI (uv) with cloud or local Chrome
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Browser agents: Jev picks each action from what is on the page while an LLM reads and plans.
yibie/awesome-jevA fast browser agent: Jev picks each action from DOM elements on the page, an LLM plans, and claims cite verifiable page quotes.
logicrw/awesome-jev-projectsBrowser agent that indexes the page into candidates for Jev to pick from, leaves planning and reading to an LLM, and cites a quote from the page for every claim.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
Which indexed page control to act on at each step.
How are answers verified?
Claims cite verbatim quotes from the page.
Do I need a cloud browser?
It is the default, but local Chrome via CDP is supported.
Related projects
Chrome extension that autofills forms from plain-text notes by having Jev match each field to the best line; skips passwords.
PluginprototypeTypeScriptPlaywright plus Jev core behind a typed SDK, persistent CLI and MCP server, with native operations needing no key and verifiable assertions.
MCP serverbetaTypeScriptRAM-efficient Rust coding-agent harness, listed as using Jev decisions for memory pruning, browser navigation and voice routing.
ServiceproductionRustMCP server and libraries that drive Chrome over CDP and hand a decision model like Jev a bounded menu of page actions.
ProjectproductionTypeScript- Jev Browser★ 335
Headless browser automation via MCP server, CLI or library: Jev picks one action per step and scores goal-met and stuck probabilities.
MCP serverbetaTypeScript Local browser agent that picks each step's operation and element with a Jev-shaped model such as a fine-tuned Laya, no cloud needed.
MCP serverbetaPython