AI-era proxy / Coding Agent ready

Bifrost

A one-stop proxy solution for the AI era. Capture real traffic, rewrite and replay requests, then hand the evidence to your Coding Agent so it can build useful skills.

Install Read Docs GitHub

$ bifrost start -d

Admin UI ready at http://127.0.0.1:8800/_bifrost/

$ bifrost traffic search --include headers,body api.example.com

24 matching requests · latest 200 OK · replay available

Install the skill bifrost install-skill

Capture real traffic bifrost traffic search --include headers,body api.example.com

Agent case

“Help me capture the login API, understand the request and response shape, then implement a reusable Coding Agent skill for refreshing sessions.”

Traffic evidence becomes rules, replay scripts, docs, and a working skill.

api.example.com reqHeaders://x-debug=1
api.example.com resDelay://350
api.example.com replay://last-success
*.internal.test passthrough://true

AI-ready Coding Agent Capture Rewrite Replay Skills

Install

Choose CLI precision or desktop comfort.

Terminal workflow

Install the CLI

Best for scripts, CI, terminals, and Coding Agents that need reproducible traffic evidence.

curl -fsSL https://raw.githubusercontent.com/bifrost-proxy/bifrost/main/install-binary.sh | bash CLI install guide

Visual workflow

Install the desktop app

Download the macOS DMG or Windows MSI, launch Bifrost, then inspect traffic and update from the app.

GitHub Releases -> Assets -> .dmg / .msi Desktop install guide

Workflow

One proxy layer between real traffic and useful AI work.

Give agents real context

Capture headers, bodies, cookies, streams, and replay material instead of describing APIs by memory.

Turn traffic into skills

Let a Coding Agent read traffic evidence, infer flows, and package the result as a reusable skill.

Replay before changing code

Convert captured requests into reproducible curl, HAR, and replay runs before the agent edits anything.

Debug the edge cases

Combine rules, scripts, values, and TLS controls when production-like behavior needs precise simulation.

Start

From zero to inspected traffic in four steps.

01 Install CLI or Desktop 02 Start the daemon 03 Add a rule for one target 04 Replay and compare