PlanBridge

Give precision feedback on every spec, plan, or message your coding agent writes

Select text & comment on rendered markdown in a local browser so your agent can act with precision

Install macOS · Linux
/bin/sh -c "$(curl -fsSL https://downloads.contextbridge.ai/cli/install.sh)"
Prefer to install manually? Read the quickstart → Works with Claude Code and Codex
Example: Claude Code's plan opening in PlanBridge.

Why it matters

Precision steering saves you from slop.

  1. 01

    Agent UX is optimized for chat, not review

    A plan, a spec, or an agent message can run hundreds of lines. Scrolling up and down to read it is tedious, and the terminal gives you no way to comment on a specific line.

  2. 02

    One text box cannot give precise feedback

    A single input box is a blunt way to respond to a spec, plan, or long message. PlanBridge lets you add inline comments, the most precise direction you can give.

  3. 03

    Less refined work gets accepted because of friction

    When iterating is tedious and hard, you stop early. What gets generated reflects the specificity of your feedback or the lack of it: great code, or slop that needs rework.

  4. 04

    Imprecision upstream costs more to fix downstream

    Refining hundreds of lines of markdown is a great investment compared to cleaning up thousands of lines of code slop.

Use cases

Three ways to use PlanBridge.

  1. 01

    Plan-mode collaboration

    PlanBridge opens automatically when your agent finishes a plan in Claude Code or Codex. Comment, then approve or send it back.

    Plan mode · Automatic Using standard hooks
    Read the docs →
  2. 02

    Precision feedback on a file

    Annotate any existing plan, spec, or doc on demand. Your comments come back to the agent.

    Claude Code
    /planbridge-open
    Codex
    $planbridge-open
    Read the docs →
  3. 03

    Annotate the last message

    Mark up the long answer your agent just wrote, then send your notes back inline.

    Claude Code
    /planbridge-last
    Codex
    $planbridge-last
    Read the docs →

How it works

How each one flows through PlanBridge.

Pick a use case to trace the loop. You comment, your agent acts, and the cheapest time to fix code is before it's written.

Coding Agent

Plan mode

Generates plan

PlanBridge

Local browser review

Coding Agent

Accept edits

Implements plan

Testimonials

Engineers love PlanBridge

“With better coding models, planning is where most of the dev time should be spent nowadays. PlanBridge makes that experience 10X better and thus increased my productivity.”
Alex B. Co-founder / CTO Fanstake
“Opening up a browser with the ability to comment inline is awesome! Highly recommend PlanBridge.”
Gururaj S. Chief Product Officer LeadLabs
“I'd recommend this for any engineer that believes planning is the new point of leverage in agentic engineering”
Larry O. Distinguished Engineer Homebound
“[PlanBridge] is very convenient. It was so hard before to make comments to the various pieces in the plan to get it adjusted. And hard to read in the CLI.”
Jacob B. VP Engineering Stellarus
“PlanBridge lets me get more quality iterations on a plan in half the time. Reviewing inline like a GitHub PR means I'm not pasting markdown into my Codex terminal like a savage.”
Alex K. Co-founder / CTO Parallax
“I think [PlanBridge] is really cool. My team is adopting it and I am going to tell everyone to start using it.”
Sean M. Director of Analytics Gamechanger
“I’ve been happy with PlanBridge. It’s so much better than trying to describe what you’re talking about in a TUI to give feedback on plans.”
Scott L. Distinguished Engineer Ibotta, Inc.
“With PlanBridge, I give a lot more feedback on plans because it’s easier to click and type a comment compared to typing in-terminal and having to scroll back and forth.”
Jesse S. Senior Software Engineer Stripe

Data Privacy

Your content stays on your machine.

PlanBridge opens a local browser review UI and returns markdown to your agent. There is no remote PlanBridge service sitting in the middle.

Read more about privacy →

PlanBridge is local-first, open source, and privacy friendly.

A local browser window gives you review ergonomics without putting PlanBridge between your agent and your content.

Local review path

  1. Markdown in

    stdin or local file

  2. contextbridge CLI

    local process

  3. Browser UI

    localhost window

  4. Markdown result

    stdout to the agent

Uninstalls easily

If PlanBridge is not for you, uninstall removes the harness entries and returns your workflow to its previous state.

contextbridge uninstall

  • Localhost UI
  • No content upload
  • No account
  • MIT licensed
  • Public source
  • Uninstalls cleanly

Works with Claude Code and Codex.

Standard plugin and hook entries for Claude Code and Codex. Other agents work too, anything that can run a command and read a structured result.

Claude Code

Plugin hook on ExitPlanMode, plus the /planbridge-open and /planbridge-last skills.

Codex

Stop hook entry, plus the $planbridge-open and $planbridge-last skills.

Setup for other agents →

Give feedback on a file manually →

Annotate the agent's last message →

Install macOS · Linux
/bin/sh -c "$(curl -fsSL https://downloads.contextbridge.ai/cli/install.sh)"

Prefer to install manually? Read the quickstart →

FAQ

Common questions.

How does PlanBridge work?

PlanBridge opens markdown in a local browser where you highlight lines and comment, then your agent acts on your feedback. There are three ways it works. Plan-mode collaboration: when your agent finishes a plan, PlanBridge opens automatically in Claude Code or Codex. Precision feedback on a file: open any markdown file with /planbridge-open (Claude Code) or $planbridge-open (Codex). Annotate the last message: mark up your agent's last reply with /planbridge-last or $planbridge-last. See the docs or the source on GitHub.

What coding agents does this work with?

Claude Code and Codex are supported with auto-detected hooks. Any coding agent that can run a command and read a structured result can use PlanBridge directly. See setup for other harnesses.

Can I open a file in PlanBridge without going through plan mode?

Yes. The planbridge-open skill opens any markdown file in the PlanBridge UI on demand (Precision feedback on a file). Use planbridge-last to annotate the agent's previous message (Annotate the last message).

Does PlanBridge take over my agent workflow?

No. PlanBridge uses standard extension points: a Claude Code plugin, a Codex Stop hook, or a manual command for other agents. It runs only when review is requested, opens a local localhost browser UI, and contextbridge uninstall removes the harness entries if you want your old workflow back.

How much does this cost?

PlanBridge is free and open source.

Who built this?

PlanBridge is built by Donnie Flood, Josh Carver, and Ben Limmer at ContextBridge AI. Three engineers who've shipped to hundreds of millions of users, building tools to make AI coding more reliable.

Why did you build this?

Coding agents are optimized for chat, not review. They give you one text input for every revision, so you can't anchor comments to specific lines. That friction wears you down, and what ships is less refined than it could be. PlanBridge gives anything your agent writes the same review experience developers expect from code review, whether it's a plan, a file you open, or the agent's last message.

Where can I request features?

Open an issue in the PlanBridge GitHub repository or join the ContextBridge Slack community.

Move faster with precision steering.

Refine first. Skip the rework.

Install macOS · Linux
/bin/sh -c "$(curl -fsSL https://downloads.contextbridge.ai/cli/install.sh)"

Prefer to install manually? Read the quickstart →