Features
What it does today, and nothing it does not.
Every line on this page is something the product does now. The lists are read from the product as the page loads. What is planned is at the end, and labelled.
Build
Agents, drawn as flows.
You draw what the agent does, step by step. A model runs the parts that need judgement; the rest follows rules you can read.
A step for every job
Conversation stages that a model runs, fixed speech, collecting details, menus, routing, conditions, tools, sub-agents, waiting for a person’s approval, and handing over to a person.
Rules, or a model’s judgement
A way out of a stage is either a rule, evaluated without a model, or a description that a model judges.
Drafts and validation
A flow is a draft until you publish it, and it is validated before it can be.
Versions and roll back
A published version never changes. Roll back to an earlier one when you need to.
A test console in the builder
Talk to an agent where you build it, before it takes a call.
Recorded and replayable
Every step of a conversation is recorded and can be replayed. A conversation survives the process that was running it.
The kinds of step
Templates to start from
The product ships with these. Their names and descriptions are its own.
Connect
Models, speech, tools, and things to know.
An agent uses the model and the speech service you choose, the tools you give it, and the documents you let it read.
Language models
Ollama on your own machine, verified here. OpenAI, Azure OpenAI, Anthropic, Google and more through one adapter, any OpenAI-compatible endpoint among them.
Marked for what has been tested
Providers that have not been verified here are built to their published APIs. Each stays marked unverified in the app until a key has been tested with it.
Speech
Calls are heard and spoken by a voice worker that takes turns, lets the caller interrupt, and reports exactly what the caller heard. Local speech recognition and voices ship with the platform.
Knowledge bases
Documents (pasted text, Markdown, a web page fetched for you, or a PDF or Word file) that an agent looks things up in before it answers a question of fact. Embeddings come from a model on your own machine or an OpenAI-style service. Search runs in Postgres, with pgvector.
Memory of callers
A stage can be allowed to note things about a caller, and to be told them the next time the same number calls. People in the workspace can see what is kept, and remove it.
Skills
Instructions an agent loads when it needs them.
Changed tools are held back
A tool from an MCP server whose description changes is held back until somebody has looked at it.
Approval before a tool runs
A tool can wait for a person’s approval. The flow carries on when they have decided.
Requests that stay outside
Requests an agent makes cannot be pointed at private addresses, and a stored key is only ever sent to the host it was entered for.
The kinds of tool
Telephony
Calls, through the PBX you already run.
The platform has a SIP gateway of its own. Nothing is installed on your PBX, which sees the gateway as one of its extensions or as a trunk.
It speaks
Key presses
A caller’s key presses are understood, so a flow can offer a keyed menu.
One address, one trunk
One PBX address belongs to one trunk on the whole platform.
Trunk status
Each trunk shows whether it is signed in and whether its PBX answers.
Opening hours
Schedules of opening hours and holidays, for flows that behave differently when you are closed.
Handover
A person, when it matters.
A flow hands a call to a person or a team by name. The agent leaves, briefs them first, or stays on the line.
What the agent does once a person has joined
And back again
The person can hand the call back to the agent.
No dead ends
Nobody answering, a busy line and a declined call each continue the flow.
Stepping in
Somebody watching a live call can have a person brought into it in any of these ways, whatever its flow was doing, or hang it up. If the person does not answer, the agent tells the caller and carries on. Who stepped in is on the audit log.
Handed over by name
A flow hands over to a name, such as Support. Who that rings is set in one place, so it can change without touching the flow.
Nothing dialled of its own accord
The platform rings only destinations a workspace set up by name, under per-trunk rules on where and how often.
Gateway
Other agents, let in on your terms.
Other AI agents, assistants and scripts can talk to the agents you build, through one MCP endpoint.
Each agent a tool
Every agent you offer appears to a client as a tool: send it a message, get its reply.
A key, or a sign-in
Clients come in with a gateway key or by signing in with OAuth 2.1.
Over Agent2Agent as well
A client that speaks the Agent2Agent protocol reaches the same agents with the same gateway key, each at an address of its own.
Held to its record
Each client is held to what its record allows: which agents, how fast, how many conversations, how many tokens a day.
A record that cannot be altered
Every request is on it.
Never instructions
What a client sends reaches an agent as what its caller said. It is never treated as instructions.
Operate
Workspaces, roles, and a record of what happened.
The parts that let more than one person, and more than one team, use the same installation.
Workspaces kept apart
One workspace’s data cannot be read from another. The database enforces it, not only the application.
API keys that carry a role
A key carries a single role, expires when you say, and is shown only once.
An audit log
A record of who changed what, with the before and after of each change.
A review of every conversation
Once a conversation is over, a model reads it and writes a summary, how the caller seemed, whether they got what they came for, a score for how well the agent did, what the call was about, and what somebody should do next. One model call, with the workspace’s own model. A workspace can switch it off.
The numbers
Volumes, outcomes, speed, cost, tool use, what the reviews add up to, and the paths conversations take through a flow.
Platform administration
A separate area for whoever runs the installation.
Runs in Docker
Everything runs in Docker on one Ubuntu server that you control.
The roles of a workspace
Start with a workspace of your own.
Create an account, connect a model, and make your first agent from a template.