Kardbrd

Give Agents Complete, Correct, Relevant Context

Store the task brief, supporting material, and working history on cards and boards. Then give people and agents the context relevant to each task.

Get Started

Project Context Needs a Maintained Record

Chat sessions often separate a task from the decisions that shaped it. A new pricing-page task may omit an approved pricing decision recorded in an earlier session, leaving the next person or agent to reconstruct why the page works as it does.

Longer context windows and saved transcripts retain more material, but retention alone does not identify what still applies. A summary can omit the correction or decision a new task depends on.

Teams need a durable record they can maintain as the project changes. It gives the next contributor a clear starting point and gives reviewers a source they can inspect.

The First Card Was Cardboard

I wrote a few ideas and questions on a piece of cardboard because it was nearby. The note gave my future self a compact record of the problem I wanted to explore.

As I researched, I added source material, constraints, and possible approaches. I gave the same note to an LLM agent, making that piece of cardboard the first card an agent could use.

That experience led to Kardbrd: a shared place where people and agents can use the same project record, track progress, and add corrections after review.

Handwritten text on cardboard: When there is nothing else to write on

A Hypothesis Worth Testing

Kardbrd is built around a practical hypothesis: an AI agent can produce better work when a person gives it complete, correct, relevant context. Each task tests that hypothesis against the result.

Useful context requires maintenance and task scope. People and agents update the record as the project changes, then provide each task only the material it needs.

A card holds one focused unit of project context. Related cards supply necessary background without forcing an agent to read the whole board. The board preserves how that work fits the project.

You set the goals and decide whether the work meets them. Kardbrd stores the record and makes it available to everyone doing or reviewing the work.

Workflow cycle diagram: Explore, Plan, Implement, Review, and back to Explore

Build, Review, and Correct

Use a repeatable cycle to build the project record and improve it as the work develops.

  1. 1

    Explore: Define the goal, study the current state, gather source material, and identify dependencies and risks.

  2. 2

    Plan: Record the constraints, decisions, and small actions required to reach the goal. Mark areas that need extra care or verification.

  3. 3

    Execute: Assign each action to a person or agent with the context needed for that part of the work.

  4. 4

    Review: Compare the result with the goal, inspect the details, verify claims, and identify gaps or incorrect assumptions.

  5. 5

    Update: Record the result, corrections, and new decisions. Start the next cycle with a more accurate project record.

Kardbrd keeps goals, decisions, experiments, artifacts, and corrections together throughout the cycle.

Three Parts of a Useful Card

A focused card combines the assignment with the material and history needed to complete it.

Task brief
The goal, scope, constraints, plan, and review criteria.
Supporting material
Research, mockups, source documents, and related work.
Working history
Decisions, progress, corrections, and outcomes from the task.

See the detailed card anatomy and workflow

Context in the Tools You Use

People can use Kardbrd from different screen sizes, and agents can access board context through supported tools.

Use a voice assistant to read or update the board while away from a screen. Create cards, move work, add comments, and continue a discussion.

During a coding session, provide the relevant card or board context to the agent working in your codebase.

Web Interface | Mobile | Claude Code | Cursor | Windsurf | ChatGPT | Claude | Gemini | Voice Assistants | Any MCP-compatible tool

Choose the context each tool receives.

Kardbrd Agent

Assign a Card. Review the Result.

Mention the agent on a card to assign the task. It works in its own workspace, posts progress as comments, and returns a result for your team to review.

Write the goal, constraints, source material, and acceptance criteria where the team already tracks the work. Link related cards when the task needs more context.

Works between sessions

Assign a task and return later to a draft, fix, or summary that is ready for review.

Handles parallel tasks

The agent can work on parallel tasks, using a separate workspace for each one.

Open source & self-hosted

Run the agent service on infrastructure you control and inspect its source code.

Keep the Next Task Grounded

Record the context, keep it current, and review whether each result meets your goal.

Get Started