Working with Tap AI
Use buyer conversations, model selection, artifacts, attachments, reusable memory, currently available integrations, guide grounding, and explicit approvals without losing your work.
5 min readUpdated
Something missing from this page? Email help@tap.co
The Tap Agent is the buyer workspace for planning and transaction preparation. Start a new chat from Home, choose a Brand when the work is brand-specific, and describe the outcome in plain language. The Agent can create or update a Plan, prepare broadcast or DOOH RFP work, compare submitted Avails, find supplier representatives, and explain how Tap works.
A conversation without a Plan stays chat-only. Once the conversation has a Plan artifact, wide screens show chat beside the artifact. On a compact screen, switch between Chat and Plan. The visible Plan remains the source of truth while the Agent is working.
Campaign team chat is a different conversation. It is shared with current Plan assignees and is not input to the Tap Agent. Read Use campaign team chat.
Use the model selector in the composer to choose the model for the current Agent conversation. The current choices are:
New conversations start with GPT-5.6 Sol. A different selection belongs to that conversation and is saved with the next message you send. It does not rewrite earlier messages or change another conversation.
Changing the model does not change your workspace access, Tap's available commands, source-grounding requirements, or approval boundaries. Choose based on the work you want to do, then review the result and the same structured actions as usual.
Recent Agent conversations appear in the navigation. Open the chat library to:
A conversation linked to a Plan follows that Plan folder. Deleting a folder moves its Plans, conversations, and subfolders up one level. It does not delete or archive their content.
Use Add files and integrations in the composer, then choose Add photos and files. Choose where the file should be reusable:
Tap uploads and parses supported files before the Agent treats their text as searchable evidence. Keep the attachment in the composer until processing finishes. If parsing fails or stays in progress, remove it, check the file, and try again. The original file and its searchable record also appear under Memories.
The Agent should cite the source it used and distinguish source facts from its own inference. Uploads are evidence, not instructions to bypass permissions or approval.
For a product how-to question, the Agent first searches the published guide, opens the relevant page in full, and cites it. This prevents a short search excerpt from standing in for the complete instructions. If the guide does not answer the question, the Agent should say so.
Open Integrations from the composer.
Currently configurable:
Currently shown as Coming soon and not connectable:
Do not tell the Agent to post to Slack, search Google Drive, use Salesforce, or open Microsoft files until the integration sheet shows that service as connectable. A service shown in the list is not available when its status is Coming soon.
Connecting a Custom MCP server inside Tap is not the same as connecting another AI client to Tap's public MCP server. Read Connect Tap to an AI client for the external-client path.
Ask the Agent to save a fact only when it should be reused. It can store:
Open Memories to review notes, edit their title or tags, link related notes, unlink them, or forget obsolete information. Replacing an outdated fact is better than saving a duplicate.
Draft work can proceed without a separate confirmation when it stays inside the workspace. The Agent pauses before consequential actions, including:
Read the consequence and acting organization before approving. Declining an approval leaves the commercial action undone.
The Agent no longer provides an email-draft command. Ask it for suggested wording in the conversation when you need copy, then send the message through the appropriate relationship or Exchange surface.
If a response fails, use the retry action on the error when it is offered. Your existing conversation and last committed Plan stay available. If a Plan update is still running, wait for its progress state to settle before editing the same fields. When a file fails to parse, retry the file from Memories instead of assuming the Agent read it.
Start a fresh chat when the work is unrelated. Continue the current chat when you want the Agent to retain its Plan, messages, and artifact context.