GroundPound.ai is part of the SurfTurf.ai family of services, built and operated by SurfTurf.AI LLC.
Feature guide
A plain-English tour of what GroundPound does.
GroundPound is a platform for running a coordinated team of AI agents that handle the work of your business — with guardrails that pause before anything irreversible. This guide walks the core feature set at a high level. For step-by-step setup once you're in, the in-app support widget and Settings panels guide you the rest of the way.
Teams, coordinators, and specialists
You don't manage one all-powerful agent. You run teams: a coordinator that triages incoming work and routes each item to the right place, plus specialists that each own a tight scope. The coordinator doesn't do the work; the specialists don't freelance outside their lane. This is the core design choice — breadth without accountability is where agents go wrong, so we trade a little speed for a lot of safety.
Getting started: describe your business
Setup is a conversation, not a form. Describe what you do in a sentence and the wizard suggests a matching template from 30+ ready-made teams, then assembles a coordinator and specialists wired for your work in about ninety seconds. See Templates for the full catalog.
Approvals: pause before the irreversible
Every agent must ask before doing anything irreversible that affects the outside world — sending an email outside your org, moving money, sending a signature, publishing to a customer-facing surface. The agent files an approval card and waits; you approve, reject, snooze, or schedule it, and the run resumes where it paused. Routine batches can be bulk-approved; high-stakes items default to deny. See Operator playbook for how to tune this to your workflow.
Knowledge: grounded in your material
Agents answer from your SOPs, contracts, and style guides — not from memory. Upload documents (up to 50 at a time, with suggested titles and tags you review), or connect Google Drive, OneDrive, Dropbox, or Box and let agents read everything with a citation on every answer. Knowledge attaches at the org, team, or agent level and inherits down.
Models: the right brain per task
You don't have to pick a model. Set your org default to Auto and the router picks the best model per task and fails over automatically if a provider is unavailable — or pin a specific model per agent, or bring your own provider key. GroundPound supports six providers (Anthropic, OpenAI, Google, xAI, Mistral, Together AI). See Technical configuration for the details.
Channels: meet customers where they are
Dashboard chat and Telegram are free; Pro adds email (your own Gmail or Microsoft 365 inbox, or a hosted mailbox on your domain), Slack, SMS, WhatsApp, Voice, and custom webhooks. Different agents can use different channels.
Workflows: extra rigor when it matters
For high-stakes output — a contract, a security review, an investor narrative — route the work through a multi-stage Workflow that puts several agents and several checks between a draft and your approval. The flagship is Adversarial Review: an evaluator, a red-team critique, and a battery of mechanical checks. See Workflows.
Atlas: watch, approve, inspect
Atlas is your command center — Triage what needs you, Operate to run and schedule, Inspect to drill into health and cost, plus a narrated activity timeline and one approvals queue across every team. See Atlas.
Extending the platform
Add tools with MCP server plugins; add house style with skill packs. Both wire in without a code change and attach at any scope. See Plugins and Integrations.
Self-healing support
A built-in support agent reads your actual setup — not a generic FAQ — and proposes a specific fix you approve in one click. A daily proactive review surfaces misconfigurations before they bite.
This is a public overview. Start free at groundpound.ai and the in-app guidance takes it from there.