Pencil / AI with product context

The AI you already trust.
The product context it needs.

Pencil brings Claude and OpenAI models into a shared product workspace. Keep requirements, features, decisions, and delivery history connected, so your team and agents can build on what you already know.

Start free See what Pencil adds

Pencil / Explore the workspaceInteractive preview · sample data

Explore Pencil

Pencil

Current sprint

Atlas Studio · Sample workspace

Sprint 12 / Onboarding

Sample tasks
Status	Priority	Assignee / Agent
In Progress	high	◎ You ▣ Dabi
blocked	medium	◎ Sam ▣ Nova
Review	medium	◎ You ▣ Aria
Done	low	◎ Sam ▣ Nova
Done	high	◎ You ▣ Dabi
Done	high	◎ Jules ▣ Dabi
Done	high	◎ Sam ▣ Dabi
Done	medium	◎ Lee
Done	medium	◎ Lee
In Progress	medium	◎ Jules ▣ Nova
Inbox	medium	◎ Lee
backlog	low	◎ Jules ▣ Aria

12 tasks · Click a task to inspect its properties.Preview only

Expand the sprint summary, switch layouts, or explore the menu. All data is fictional and interactions stay in this preview. Pencil’s default dark theme is shown; navigation and fields are configurable in your workspace.

Why Pencil / Beyond the conversation

Same model power.
A shared memory for your product.

Claude, ChatGPT, and other AI assistants are useful places to think, write, and build. Pencil adds a structured home for the product work around those conversations: what you need, why it matters, who owns it, and what happened next.

What’s running behind the scenes?

Pencil can use Claude from Anthropic and GPT models from OpenAI, the company behind ChatGPT. You get the capabilities of the model you choose, with Pencil organizing the context and workflow around it. Available models and billing depend on your setup.

A conversation on its own, and a product workspace around it. Scroll horizontally on smaller screens to compare.
The work	AI chat on its own	What Pencil adds
Product needs	Describe the feature and supply its background in your conversation.	Track ideas, requirements, and features as shared product records.
Decisions + history	Bring the relevant prior discussions and decisions into the work.	Keep decisions, comments, files, and memories accessible alongside the work.
Team handoffs	Share the conversation and explain who should do what next.	Give tasks human and agent owners, priorities, status, and review requirements.
From plan to delivery	Carry the result into your planning and delivery tools.	Connect product plans to tasks, agent outputs, GitHub work, and deployments.
Visibility	Review the answer or generated code in the conversation.	Inspect task progress, outputs, and usage by agent, model, project, and task.

AI assistants may also offer memory, projects, and integrations. This comparison describes workflows; Pencil’s focus is a shared, structured product record across people, agents, and tools.

Claude + Pencil / Connected through MCP

Stay in Claude.
Bring Pencil into the build.

Use Pencil directly, or access it from Claude through Pencil’s MCP server. Add the Pencil skill to give Claude your workflow instructions, then ask it to read the relevant requirements and decisions before it builds.

Connect your workspace.Connect Pencil’s MCP server and authorize access to the right workspace.
Give Claude the Pencil skill.The skill guides the workflow; the MCP connection gives Claude access to Pencil’s tools and records.
Read, build, and report back.Ask Claude to check context first, surface open questions, and record the result in Pencil with the permissions you grant.

Connect Pencil to Claude

Example workflow / Team invitations

You, in Claude

“Build team invitations. First, check Pencil for the requirements and decisions we’ve made about this feature. Use that context in the implementation, flag anything unresolved, and save a summary of the work back to the task.”

01 / Read from Pencil

Start with the agreed scope

Requirement: admins can invite teammates by email.
Decision: invitations expire after seven days.
Open question: can guests invite other guests?

02 / Build with context

Let the decisions shape the work

Use the invitation rules in the implementation. Ask for a decision on guest permissions before expanding the scope.

03 / Save the handoff

Leave the next person a starting point

Record the implementation summary, test results, and remaining questions on the Pencil task.

Illustrative example. Claude needs authorized MCP access and development tools to implement code; the skill alone does not connect your workspace.

People + agents / Shared priorities

A workspace that keeps
everyone on the same page.

Clear ownership

Assign a person and an agent to the same task. Keep responsibility visible as work moves across the board.

Reviewable work

Follow progress, inspect outputs, and review agent contributions in the context of the task.

Lasting context

Keep decisions, files, and agent memory connected to the work they inform.

Your workspace, your level of structure

Start with what you need. Add more as you grow.

Start with shared tasks, clear owners, and collaboration. Turn on more tools when your team needs them—not before.

Start simple

A shared board. A clear next step.

TasksOwnersCollaboration

Add a little structure

Bring in the tools that fit your work.

SprintsClientsSub-tasks

Connect the bigger picture

Expand when your team is ready.

RoadmapsProduct specsReports

Three short steps. No account needed.Explore all features

Tasks / Humans + agents

An owner.
An agent.
A clear next step.

Pencil keeps status, priority, human assignment, and agent assignment with the task. Open a card to inspect the brief and the work behind it.

Delegate. Inspect. Review.

Use checklists and agent outputs to understand what’s ready. Configure plan approval and review requirements for the work that needs them.

Explore all features

MC-132Task detail · Sample data

Review onboarding copy

Keep the welcome flow focused: introduce the workspace, find a first task, and make the next step clear.

Properties

Status: Review
Priority: medium
Agent: ▣ Aria
Assignee: ◎ You
Estimation: 3pt

Checklist Checklist 3/3

☑Define the first step
☑Link the supporting context
☑Review with the team

Agent output

Example output: onboarding-copy.md
Saved against this task for review.

Memory / Context that stays

Give the next task
a better starting point.

Browse an agent’s memories and files side by side. Keep facts, decisions, and lessons accessible, and inspect the context your agents use.

Agent memories · MEMORY.md · lessons.md

Pencil

Memory

Agent · Nova (Head Agent)

Memories

decision

Keep onboarding to three steps.

onboarding · 2h ago

fact

The launch brief is attached to MC-129.

launch · 1h ago

Files · Nova

# Project context

Keep onboarding short.
Link work to its brief.

## Next
Review the launch copy.

Sample agent memories and files. No live workspace connection.

Costs / Usage you can understand

See what the work
actually consumes.

Track usage by agent, model, project, and task. Pencil separates equivalent token usage from billed spend, so flat-rate and local runs don’t disappear from the picture.

Usage trends · Cost breakdowns · Model routing

Pencil

Costs

30d · Usage

Spend (equiv.)$24.80$8.20 billed

Tokens1.2M64 events

Top agentNovaExample usage

Cost by agent

▣ Nova$12.40

▣ Dabi$8.68

▣ Aria$3.72

Illustrative data · Equivalent usage can differ from billed spend.

Integrations

Keep your tools.
Connect the work.

Bring conversations, documents, and code into the same workspace.

SlackConversations into tasks

GitHubCode linked to work

LinearConnected planning

NotionIdeas and documents

MeetingsNotes into next steps

MCP + API

Your clients.
The same workspace.

Connect external AI clients through hosted MCP or a paired local gateway. Use the REST API for custom tools and structured integrations.

Read MCP docs API reference

Hosted MCP: /api/mcp
Local gateway: pencil-gateway mcp
REST API: /api/v1

Built in Pencil. Building in the open.Follow our roadmap

Your next project starts here

Keep your AI.
Give your product a memory.

Start freeIncludes $5 in trial credits.

The AI you already trust.The product context it needs.

Current sprint

Same model power.A shared memory for your product.

Stay in Claude.Bring Pencil into the build.

Start with the agreed scope

Let the decisions shape the work

Leave the next person a starting point

A workspace that keepseveryone on the same page.

Clear ownership

Reviewable work

Lasting context

Start with what you need. Add more as you grow.

Start simple

Add a little structure

Connect the bigger picture

An owner.An agent.A clear next step.

Delegate. Inspect. Review.

Review onboarding copy

Give the next taska better starting point.

Memory

Memories

Files · Nova

See what the workactually consumes.

Costs

Cost by agent

Keep your tools.Connect the work.

Your clients.The same workspace.

Keep your AI.Give your product a memory.

The AI you already trust.
The product context it needs.

Same model power.
A shared memory for your product.

Stay in Claude.
Bring Pencil into the build.

A workspace that keeps
everyone on the same page.

An owner.
An agent.
A clear next step.

Give the next task
a better starting point.

See what the work
actually consumes.

Keep your tools.
Connect the work.

Your clients.
The same workspace.

Keep your AI.
Give your product a memory.