Summary
A Reply Agent is an Alta assistant that handles inbound replies on a campaign — it reads each reply, decides what to do, and either replies on its own or labels the outcome so a human can take it from there. The builder has two steps: Goal (what the agent should do, with what tools, in what voice) and Outcome (how it classifies the conversation when it's done).
Who this is for: Anyone setting up campaign reply automation in Alta — typically the campaign owner or RevOps.
Before you start
You'll need permission to create assistants in your Alta workspace.
If you want the agent to book meetings or check availability, the relevant rep should already have calendar tools connected (Settings → Katie's Reps).
If you want the agent to push or read CRM data, your workspace CRM should already be in a callable state (Settings → Integrations).
Have your knowledge base files / URLs ready if you want the agent to ground its replies in your own content.
Step 1 — Create a Reply Agent
Go to Assistants in the Alta sidebar and create a new Reply Agent.
Name it from the top bar (placeholder: Reply Agent Name). The stepper between Goal and Outcome lives in the same top bar.
The top right has two save buttons:
Save — saves and stays on the current step.
Save & Continue — saves and advances to the next step.
If you try to jump between steps with unsaved changes, Alta opens the Leave Page dialog. Save first, or accept that your edits will be reset.
Step 2 — Fill in the Goal step
This is the "Context and clarification" screen. Five sections, all expanded by default.
Goal
"Describe the goal for this assistant. i.e. BANT qualification, book meetings, etc."
In the prompt text area, write what the agent should do. Use the variable autocomplete to pull in supported prompt variables (Reply Agent variables — for example, prospect name, company, or campaign context).
The prompt is capped at 6,000 characters. Keep it short and behavioral — "qualify against BANT and book a meeting if interested" beats a 5-page playbook.
Content and sources
"Provide knowledge base files and sources to help your assistant respond with relevant information."
Attach any combination of knowledge base files and URLs. The agent will draw from these when its prompt references them.
Tools
"Guide your assistant's tools to complete the task." Three controls:
Use rep linked tools (toggle) — "Lets the assistant use calendar tools tied to the rep (book meeting and check availability). Workspace tools below stay separate." Turn this on for any agent that needs to book or check meetings on the rep's calendar.
Built-in Alta tools (multi-select) — pre-shipped tools that ride on top of your existing connections. Options whose prerequisites aren't met (for example, CRM-dependent tools when no CRM is callable) are still shown but appear unavailable.
Workspace tools (multi-select) — "HTTP, MCP, and other tools from your workspace. Calendar scheduling integrations belong on the rep or via rep-linked tools above." The Manage tools link in this section opens Settings → Tools in a new tab if you need to add new ones.
Agent's tone of voice & Response length
"Choose the tone of voice and response length for your assistant."
Agent's tone of voice — pick one: Friendly, Neutral, Matter-of-fact, Professional, or Humorous.
Response Length — pick one: Concise, Standard, or Thorough.
Model
"Choose the model to use for this assistant." Pick the LLM model the agent should run on from the Select LLM model dropdown.
Step 3 — Fill in the Outcome step
"Describe the outcome for this assistant (e.g. For lead qualification: Qualified, Unqualified, Nurture, etc.)."
This is where you turn the agent into a structured classifier. Each outcome has a Status, a Prompt describing when to assign it, and a color used in the inbox.
Click Add status. Each click adds a new row.
In Enter status, type the short label you want the conversation tagged with (e.g. Qualified, Nurture, Not a fit).
In Describe the criteria for this status, write what has to be true for the agent to pick this outcome. Be concrete — "prospect confirmed budget and timing" beats "prospect seems interested".
Repeat for every outcome you want available. Common starter sets: Qualified / Unqualified / Nurture or Booked / No-show / Lost.
Save the step.
Tips and common pitfalls
Calendar tools belong on the rep, not in Workspace tools. If the agent needs to book or check availability, flip Use rep linked tools on. Don't try to add the calendar tool from the Workspace tools list — Alta filters those out by design.
Built-in tools may be greyed out. Some require an integration to be callable first (notably CRM). If a tool looks unavailable, check Settings → Integrations to make sure the relevant integration is connected and active.
Outcomes should be mutually exclusive. Overlapping criteria ("Qualified" and "Booked" with the same description) confuse the classifier. Write each criteria so only one can be true at a time.
6,000 characters is plenty. If you're pushing the limit on the Goal prompt, move details into knowledge base files instead — the agent will retrieve them when relevant.
Pick the tone with your inbox audience in mind. Humorous can land well for self-serve SaaS reply threads but feels off on a regulated industry outreach. Matter-of-fact + Concise is the safest default.
Save before you switch steps. Unsaved Goal edits will be reset if you click into Outcome without saving.
Related
