Ask across the family: one assistant that answers and acts across the products

What the planned family assistant will do: answer a question from every product with the source named, and make changes under your own identity, with each change approved in PromptCompute.

Ask across the family will be a family assistant: one chat that answers questions from every product, and can make changes in them for you. It will configure accounts, connected services and settings, acting under the identity of the person or agent who is signed in. Every change will go through a PromptCompute approval, and every answer will name the product it came from.

It is planned, not built. This guide describes what it will do, the rules it will follow, and the pieces that exist today.

The problem

A simple question means opening several products. Did we promise this customer SSO, and has it shipped? The promise is in CommitOps, the deal is in Sales as Code, and the shipped change is in RepoOps. Someone opens three tabs, reads three screens, and puts the answer together in their head.

Changes are the same. Connecting a service, adding a teammate or changing an approval rule means finding the right settings page in the right product, each with its own sign-in. An agent has it worse. Today it needs a separate key for each product it calls.

How the assistant will solve it

You will ask once. The assistant will read the facts each product publishes and answer, naming the product and the date behind each part of the answer. When two products disagree, it will show both answers, each with its owner, and will not merge them into one. You will see the disagreement instead of a guess that hides it.

When you ask for a change, the assistant will draft the exact action and its arguments and send it to PromptCompute for approval. A person decides. The change will run only after that, under your identity, so the assistant will be able to do only what you could do yourself in that product. The decision is signed, like every PromptCompute decision, so the record will show who asked, who approved and what changed.

Step by step

The assistant is planned. Present tense below means a piece works in its product today; “will” means it is still to build.

For a question:

1. You or your agent will ask in the chat.
2. The assistant will read each product’s facts. RepoOps already answers from its brain with ask_brain, PromptCompute already answers owner questions from its family fact lookup, and AI Source of Truth already serves the family facts list.
3. The assistant will answer, naming the product and the date of each fact. Reading changes nothing, so a question will need no approval.

For a change:

1. You will ask for the change, such as connecting a CRM to Sales as Code.
2. The assistant will draft the exact action and arguments in the product that owns them.
3. PromptCompute holds the action for a person. The approver sees the exact action before it runs, in Slack, email or the web, and the agent picks up where it left off once they decide.
4. The product will make the change under the identity of whoever asked, and the assistant will report what changed.

What the assistant will report

The plan sets two measures:

• Answers with a named source. The target is 100%.
• Questions answered without opening a product.

For changes there is a rule instead of a target. Every change will carry its PromptCompute approval, and a change with no approval will not run.

Why it matters to a founder or an engineering lead

People and agents will stop guessing. A teammate who asks whether a promise shipped will get the answer with its owner and date, and can trace it. An agent that today juggles a key per product will work through one assistant, under one identity, with each change waiting for a person.

Two rules are what will let you give the assistant to the whole team. It will never do more than the person using it could do, and nothing it changes will skip an approval.

How to set it up

The assistant needs no connection outside the family. It reads and acts through the products you have already connected.

When it ships, setup will be short:

1. Sign in with PromptReports ID. The assistant will act as you.
2. Connect the products you want it to read and act in.
3. Set up approver groups in PromptCompute, so each kind of change reaches the right person. Every change will go through approval whatever the policy says; the groups decide who approves it.
4. For an agent, register it in PromptCompute so it has its own identity. The ID’s agent token will name the person who started the agent (sub) and the agent and its version (act), so a change made by an agent is recorded as that agent, acting for that person.

The assistant will live in two places: the signed-in app at app.promptreports.ai, and a tool on the family MCP endpoint, so an agent can ask it too.

Every plan includes every workflow. Plans are per user per month: $25 for 100 workflow requests, $50 for 500, and Enterprise above 500. There is no free tier. Each question to the assistant will count as one workflow request, and so will each change it makes. The allowance is pooled across your organization: the number of users times the plan’s requests. Every model call the assistant makes will use your own model provider key, and PromptReports never pays for model calls or marks them up.

What is built today and what is planned

The workflows page lists Ask across the family as planned. We decided on 2026-11-10 to make it a family assistant that can act, not only answer.

In the products today:

• RepoOps answers questions from its brain with ask_brain.
• PromptCompute has an owner fact lookup for the family.
• AI Source of Truth serves the family facts list. It is empty today, because no org has opted in to sharing public facts yet, so an assistant reading it now would have nothing to answer from.
• PromptCompute holds an agent’s action for a person and signs each decision.
• PromptReports ID is live at accounts.promptreports.ai. Signing in to each product through it is built and switched off, and its agent token exchange is built and off until its key is set.

Planned:

• The assistant itself: the chat, reading every product’s facts with their owners, and drafting changes.
• Every product accepting the PromptReports ID agent token, so the assistant can act under one identity in each of them.
• The same Ask in the customer portal for people who do not use RepoOps.
• The workflow runner the other workflows share. No runner exists today. The plan is a durable runner service in the PromptCompute stack, with one run id across products, approvals as steps, and each run metered as one workflow request.

Read more

• Every workflow and its state
• Pricing
• One sign-in, one approval, one record across the family
• RepoOps, PromptCompute and AI Source of Truth