Our Tools

Software we built because the work needed it

Each of these started as a problem we hit on an engagement and couldn't solve with something we could buy. We use them ourselves before any client sees them, which is the only reason we trust them enough to offer them.

Solutioner.aiSkillhouseAtlasPrompt Manager
Live

Solutioner.ai

The pipeline for AI work.

Most companies collect AI ideas in a spreadsheet, a Slack channel, and three people's heads. The ideas are usually good. What's missing is a way to say which ones are worth doing, in what order, and what became of the last dozen.

Solutioner turns each idea into a solution brief: the process it touches, what that process costs today, what it would save, and how complex the build is. Briefs move across a board from draft to delivered, and the numbers roll up, so a leadership team can look at the whole portfolio instead of a handful of anecdotes.

/o/dunder-mifflin/dashboard

The business case comes first

Each brief records annual process cost, projected savings, time reduction, and quality gain. You know what a solution is worth before anyone writes a line of code, and you can compare two of them honestly.

Three ways to get an idea in

A guided form for people who want structure, a quick submit for people who don't, and a conversation with Sully, the built-in assistant, for people who would rather talk it through than fill anything out.

Automatic leveling

Briefs are scored for complexity, so a two-hour automation and a six-month platform build don't sit in the same undifferentiated queue competing for the same attention.

The program, not just the pipeline

Wins tracking with actual realized savings, periodic operator snapshots that show where your team really stands, live lightning rounds during sessions, and a workshop schedule.

Who it's for: Companies running an AI program across more than one department.

Visit solutioner.ai →
Live

Skillhouse

Governance for the skills your AI reads.

Anthropic solved distribution. Connect a GitHub repository to your Claude organization and everyone has your skills within about thirty minutes. Management was left to you, and it locks out exactly the people who should be writing these files. The sales lead who knows what a good discovery call sounds like has no GitHub account, and shouldn't need one.

Skillhouse sits between your people and that repository. It knows your departments, who belongs to them, and who can approve changes for each one. An employee works on a skill with Claude in plain conversation, Skillhouse checks the draft before it is ever submitted, and the department's approver sees a summary, the risk flags, and a side-by-side diff. On approval it commits through a GitHub App, or opens a pull request instead if your IT team wants a gate.

/o/dunder-mifflin/proposals

Proposals happen inside Claude

Claude fetches the live skill through Skillhouse, works the change out with the requester in conversation, and submits it. No repo access, no branches, no ticket sitting in someone else's queue.

Checked before a human sees it

Every draft is linted and reviewed by Claude first, producing a plain-English summary and risk flags, such as a skill that would send client data to an outside system. The requester fixes problems before an approver ever spends time on them.

Approval that matches your org

Proposals route to the approvers for that department. They can approve and publish, request changes with a note, edit before approving, or reject. A base-drift warning fires if the skill moved underneath the proposal.

A real audit trail

Every change is a signed commit. Every version is kept, any skill can be rolled back, and the log records who proposed, who approved, and when it went live.

Who it's for: Any company whose Claude skills live in a repo that one person has become the bottleneck for.

Read why we built it →
In development

Atlas

A living map of how the company actually runs.

The standard deliverable for an operations assessment is a deck. It is accurate the week it lands and stale by the next quarter, because nothing about a slide invites anyone to update it. The knowledge it captured goes back to where it came from: who hands what to whom, which step everyone quietly dreads, and which spreadsheet the entire process depends on.

Atlas replaces that deck with a map the client keeps. Our consultants build it during an embed, either on a canvas or by talking to Claude, and every person, workflow, step, decision, and system becomes a typed record with provenance and a verification state. Bottlenecks, manual work, tribal knowledge, and failure points attach to the exact steps where they occur, which is precisely where the opportunities are.

/w/dunder-mifflin

Captured by talking, not by drawing

A consultant tells Claude how the rights team works and the assistant extracts the people, steps, handoffs, and systems, writing them to the map as unverified until a human confirms them.

A typed graph, not a whiteboard

Eleven node types and seventeen kinds of relationship, each a structured record rather than a sticky note. That is what makes the map queryable, and what lets it be checked for what's missing.

The problems sit where they happen

A bottleneck badge lives on the approval step that causes it. Manual work lives on the step being re-keyed by hand. You read the map and the opportunities are already located.

It survives the engagement

Gadoci provisions and populates the workspace during the embed. Afterward the client's own maintainers keep it current, which is the whole point. The asset outlives the consultant.

Who it's for: Clients in an AI embed engagement who want the assessment to still be true a year from now.

Ask about early access →
Internal tool

Prompt Manager

A queue between a consultant and Claude Code.

When you are running work across a dozen projects, ideas arrive faster than anyone can act on them. The usual responses all fail in their own way. Start one and the rest evaporate. Dump everything into a single long prompt and the model splits its attention and does each piece a little worse. Stash them in a notes app and within a week they are sprawled across a dozen files, out of order and disconnected from the project they belonged to.

Prompt Manager gives every project a board. Prompts land in Backlog the moment you think of them, get dispatched into a live terminal session one at a time so each one gets clean context, and finish in Done with the result attached. Because the board belongs to the project rather than to any one terminal window, the history travels with it.

/projects/dm-order-portal

Capture without breaking focus

Type the idea as a sentence and keep working. You stop holding five things in your head, and you stop interrupting the task in front of you to service each one as it arrives.

One job at a time, in order

The queue runs sequentially. Stack five jobs and walk away, and they run one after another with clean context each time, which is why the output is better than one prompt carrying ten unrelated asks.

Context that outlives the session

A Claude session forgets when it ends. A project doesn't. Open one cold after a few days and a one-click recap tells you where things stand, drawn from what actually finished.

Works from your phone

Queue work, reorder it, and see which sessions are running and which are waiting on a human answer, all from away from the desk.

Who it's for: Our own consultants. This one runs the shop rather than being something we sell.

Read how it works →

Want one of these inside your company?

If you're interested in deploying one of our tools inside your company, get in touch. We'll tell you honestly whether it fits, and what it would take.

Get in Touch