Tools it can call
Every action is a named tool with a written description: look up an account, transfer a call, start a workflow. Your team can read the whole list, and the agent acts only through the tools on it.
AI agents that ask before they act.
An agent is a model that can use tools: look something up, transfer a call, start a workflow. In what we build, a person approves anything consequential and every step is recorded, on infrastructure you control.
What we mean by an agent.
The word gets used for almost anything. Ours has three parts, and we design all three before an agent touches a live system.
A person for consequential actions
Reading and routing can run on their own. Anything that changes a record, sends a message, or moves money waits for a named person to approve it.
A record of what happened
What was asked, which tool ran with what inputs, what came back, and who approved it, kept on your infrastructure where your compliance team can review it.
Where people step in and how we watch a system once it is live apply to agents like everything else we build. They are covered in people in the loop and monitoring on our approach page.
Where agents run today.
Agent capabilities in our own products: what is live, what is partly built, and what is planned.
Call transfer in Chat411
During a live call, the voice agent calls a transfer tool when the caller asks for a person, connects them to that extension, and leaves the call. It runs on Asterisk and FreePBX.
Live today
Lookups and actions in your systems
Chat411’s tool framework is built so each new action is another named tool. CRM and scheduling lookups come next, and a person approves anything that changes a record. Today, transfer is the only tool.
Groundwork built
Agent workflows in WolfPackAI
n8n runs inside WolfPackAI, our alpha AI gateway, for AI-driven workflows and webhooks into your systems. Full n8n integration is still in progress.
Groundwork built
Role-based access in WolfPackAI
Control over which people and teams can use which models and workflows. It is on the WolfPackAI roadmap for the beta release.
Planned
A saved record of every call
The transcript, a short summary, and where the call went, stored for review. Today, operators can follow the transcript while the call is live.
Planned
Outbound calls with oversight
Appointment confirmations and follow-up calls placed by the agent, each one logged so your team can check it.
Planned
Chat411 has two supported voice paths. Local path: a local model, media that can be encrypted, and no required PSTN connection. Audio can stay on your network. Cloud realtime path: caller audio streams to a provider, xAI by default. Call control stays on your PBX in both paths.
The full voice roadmap is on the PalmConnect page.
Agents built into the systems you run.
The most useful agents work in your systems, not ours. As part of an engagement, we build the tools an agent calls, the screens where people approve its actions, and the records it leaves behind.
- CRM and ERP lookups and updates, with approval before anything is written
- Asterisk and FreePBX call handling, from a team that has integrated with Asterisk for more than a decade
- Document systems: agents that find, compare, and summarize files, often backed by private RAG
- Workflow automation with n8n and webhooks, inside WolfPackAI or next to the tools you already use
- Fast checks before an agent acts: a System One model can screen a request or choose a route in milliseconds
Which model runs the agent
We test candidates on your own tasks first, including how reliably each one picks the right tool. The winner might be an open-weight model on hardware we build for the workflow, a commercial model through a gateway, or a mix of both.
When not to use one
If a fixed workflow or a simple rule does the job, we will say so. Agents earn their place where requests vary too much to script.
Questions about AI agents
What is agentic AI?
An AI model that can take actions by calling tools, not only write text. In the systems we build, each tool is named and reviewable, a person approves consequential actions, and every step is recorded.
Do your agents act without a person?
Only for low-risk steps, such as a lookup, or transferring a caller who asked for a person. We design anything that changes a record, sends a message, or moves money to wait for a named person to approve it. In Chat411, transfer is live today and approvals are groundwork built.
Can AI agents run on-premises?
Yes. The agent service, tools, and records can run on your infrastructure, with a local model on your hardware. Chat411 also supports a local voice path: media can be encrypted, PSTN is not required, and audio can stay on your network. Its separate cloud realtime path sends caller audio to a provider, xAI by default.
Can an AI agent work with our Asterisk or FreePBX phone system?
Yes. Chat411 answers calls on Asterisk and FreePBX through the Asterisk REST Interface and transfers callers to extensions today. Your numbers, carriers, and routing stay where they are. What your PBX needs.
Which models do you use for agents?
Whichever tests best on your tasks, including how reliably it picks the right tool. That can be an open-weight model on your hardware, a commercial model through a gateway, or a small decision model that routes requests before a larger one acts. Open model deployment.
Bring one workflow you would hand to an agent.
We will map the tools it would need, where a person approves, and what we would record.