FAQ: AI Agent, Memory, Skills & Playbooks

AI Agent

Which plans and user roles have access to the AI Agent?

Every Topo user has access to Topo's AI Agent.

Do agent actions consume credits?

No. Using the conversational AI Agent does not consume Topo credits. Only enriching leads does.

Is chat history shared with the team or personal only?

Chat history is not shared with the team. It is personal to you.

Can the agent do things without asking me first?

It depends on Approval mode in the composer. Manual, the default, asks before every action that sends outreach, spends credits or changes your setup. Auto runs on its own and asks only before the actions it judges risky. Choosing how many leads to import, and which enrichments to run, still asks in either mode.

Can I stop the agent while it is working?

Yes, you can stop the agent mid-run.

Can I undo something the agent has already done?

No. The agent can only perform the same actions you can, so it is bound by the same limits: if you cannot reverse an action in the app, un-enriching a list for example, neither can the agent.

A run failed. Do I have to type my request again?

No. Click Retry under the failed run and the agent picks it up again.

Can I use the agent from Claude or ChatGPT?

Yes. Connect Topo to Claude, ChatGPT, Cursor or Codex from Settings → Topo MCP. See 📄 Using Topo from Claude, ChatGPT, Cursor or Codex.

Does the agent work in French?

Yes. Write to it in the language you work in and it answers in that language.

Can I set it running overnight, unattended?

Yes. Agent schedules run the agent in the background, once at a set time or on a repeat. When the run finishes, the result can go to Slack, by email, or stay in Topo for you. Set one up in Settings → Agent Schedules.

Can a scheduled result go to someone who doesn't use Topo?

Yes. Choose Email as the delivery and list up to 10 people, with or without a Topo account. They get the full answer in the email.

It will not answer my question.

The agent is scoped to prospecting, outbound, go-to-market and support, and declines anything outside that on purpose.

I mentioned a lead and it refused to touch it.

Working as intended. Mentioning a record means you are talking about it, not instructing the agent to change it. Ask explicitly, for example "unenroll this lead," or mention the record together with a skill so it becomes that skill's input.

My conversation disappeared.

Coming back after more than eight hours starts a new conversation. The old one is still in your history. In a very long conversation, earlier turns are summarized rather than kept word for word, so small details can fade.

It stopped mid-answer.

A single request has a time ceiling, and a very heavy multi-step ask can hit it. Split the work into two. The stop button also cancels on purpose. A cancelled card reads "Cancelled".

A chip is greyed out and struck through.

That means the record is no longer available, either because it was deleted or because you can no longer see it.

Can I use the agent in Slack?

Yes. Mention @Topo in a channel, or message the Topo app directly. It is the same agent as in the app. The Slack integration has to be connected first.

The agent cannot see one of my tools.

Check Connectors. Call recorders and knowledge bases need connecting before the agent can read them, and a connector stops working if its authorisation expires or is revoked. Your CRM is different: it's integrated natively and needs no connector.

Agent Memory

How do I edit or delete an existing memory?

Scroll to the right of the memories table, click the three dots in the Actions column, then select Delete.

Can an Episodic memory be promoted to Durable?

No. Once a memory is saved, it cannot move from Episodic to Durable, or the other way round.

Who can edit an Organization-scoped memory?

Anyone in your organisation. Treat organisation memories as shared team settings.

Is there a cap on the number of memories?

No, there is currently no cap.

It said it remembered my ICP but Agent Memory is empty.

By design. ICP, value proposition, audiences, competitors and positioning are saved to Organization Context, not to Agent Memory.

What I told it a few minutes ago isn't showing up in memory.

Memories are written a little while after the conversation goes quiet. Say "remember this" to have it recorded on the spot.

How does Agent Memory relate to Organization Context?

You do not need to repeat in Agent Memory anything already stored in Organization Context. Use Agent Memory for everything that is not in there: your outbound routines, tone, preferences and similar.

Skills

Can a skill be edited, duplicated or deleted after creation?

Yes, all three, from the … menu on the skill card. This covers the built-in Topo skills as well as the ones you build.

Will Topo change my skills in the background?

No. A skill changes when you edit it, or when you ask the AI Agent to update it.

Can I force the agent to use a specific skill?

Yes. Type @ in the chat and mention the skill by name. Otherwise the agent picks the skill itself, based on each skill's Description, as long as Auto-discovery is on.

What can @ Insert reference point to?

Lists, sequences, playbooks, lead searches and other skills.

Is there a cap on the number of skills, or on skill length?

No cap on either.

It ignored my skill.

Three things to check: the skill is archived, its auto-discovery is off (then it must be @-mentioned), or the message matched more than three skills (only three apply, and the agent says which ones it followed).

Playbooks (coming soon)

Will a playbook contact leads on its own?

Only if you build it that way. Nothing is sent unless a step in the flow sends it. If you want a human check before anything goes out, add an Approval step and the run pauses until someone approves.

Do playbook runs consume credits?

Only if the playbook enriches leads.

Can I duplicate a playbook?

No. Install the template again, or build a new one, if you need a variant.

Who can publish a playbook?

Any member of your workspace.

Do Slack notifications need anything set up first?

Yes. Slack notifications only work once the Slack integration is connected in Settings > Integrations.

Are the templates the same in every workspace?

Yes. All 19 templates are identical across workspaces.

Does publishing turn a playbook on?

No. Publishing saves a version. A playbook that has never been published shows Draft. A published playbook that is not running shows Off, and the editor pill reads Published · off. Use the On toggle, or Turn on, to start it. If it is not already running, the Playbook published toast says it won't run until you turn it on, and offers Turn on.

How do I test a playbook?

In the editor, on the same canvas. A bar at the bottom shows the selected test case and Run test, which runs every case. Add a case from that bar: Replay a past run, Generate with AI, or Start from scratch. When the trigger is a signal, the menu also has From a signal. It copies a recent imported occurrence that matches the trigger, and the case is described as From a real signal. On a signal's Imports log, an imported event has Test in a playbook. Open a step and use its Output tab for During tests and Test this step. There is no separate Test mode.

Can one playbook start from several signals?

Yes. On a signal trigger, Specific signals lets you pick more than one signal you already set up. The playbook starts when any one of them fires. The start node then reads One of N specific signals fires.