05 / Products
Chat/Agentic Chat Interface
A conversation on top of your agents.
Chat is a full-page conversational interface to Linear Horizon Platform. It streams answers as they are generated, can show the working behind them, and supports hands-free voice — while the agent system does the actual work.
Interface
A message is routed to an agent, worked with tools, and streamed back with its working visible.
The conversation is the visible layer.
Chat is the visible layer over an agent system. It talks to Platform through a same-origin backend, so the browser only ever speaks to one trusted origin and enterprise sign-in and permissions apply as they should.
It is deliberately more than a reply box. A picker lets people choose the right agent for the task; a "show working" view reveals the tool calls, tool results and sub-agent activity behind an answer; and responses stream token by token so the conversation feels immediate.
A conversation mode adds speech: utterances are transcribed with recent context, and replies are spoken sentence by sentence as they stream — so Chat can act as a hands-free scribe or copilot as well as a text interface.
A basic chatbot replies. Chat is built to sit in front of an agent system: it selects the right agent, streams the answer as it is produced, and can show the tools and sub-agents that produced it — so people can trust what they are reading.
How a turn works
Ask a question. Watch the work. Read the answer.
A message is sent to the selected agent through a same-origin backend. The agent may call tools and retrieve knowledge; intermediate steps appear as working, and the final answer streams back token by token into the live message.
- 01
Ask
- 02
Route to agent
- 03
Tools / RAG
- 04
Stream tokens
- 05
Show working
- 06
Answer
What Chat provides
- 01
Agent selection
Choose the right agent for the task from the catalogue your Platform deployment exposes.
- 02
Token streaming
Answers stream into the live message as they are generated, so a reply starts before it is finished.
- 03
Show working
Reveal the tool calls, tool results and sub-agent activity behind an answer when transparency matters.
- 04
Conversation mode
Speak instead of type: utterances are transcribed with context and replies are spoken as they stream.
- 05
Same-origin backend
The browser talks only to one trusted origin, which proxies to the private agent service — no cross-origin exposure.
- 06
Enterprise sign-in
Authenticate through your identity provider so conversations run as the person asking, with their permissions.
Use cases
Where Chat is used.
01
Internal assistant
The problem
Teams have agents that can retrieve knowledge and take actions, but no interface a non-technical colleague can actually use.
How it is addressed
Chat gives them a front door: pick an agent, ask in plain language, and see the working when they need to trust the answer.
02
Customer-facing chat
The problem
A public assistant has to stream quickly, stay grounded and run behind proper authentication — not leak a backend to the browser.
How it is addressed
Chat serves the UI and proxies every call to the private service from the same origin, so streaming, grounding and sign-in are handled in one place.
03
Hands-free copilot
The problem
Some work happens away from the keyboard, where typing a request and reading a screen is not practical.
How it is addressed
Conversation mode lets a person speak requests and hear replies as they stream, using Chat as a scribe or copilot.
04
Reference implementation
The problem
Teams building their own frontend need to see how a browser should talk to an agent service, end to end.
How it is addressed
Chat is a working example of agent selection, streaming, visible working and a same-origin proxy that a product team can learn from.
Limits worth being clear about
An interface to an agent system, not the system itself.
- 01
Chat is the interface, not the intelligence. Platform and its agents do the work; Chat presents it.
- 02
It is not designed to disguise automation as a human. Where appropriate, an automated conversation should be identifiable as one.
- 03
What an agent can answer or do depends on the knowledge, tools and permissions configured behind it.
- 04
Voice uses the browser’s speech capabilities, so quality and availability vary by device and environment.
Enterprise deployment
These products are delivered as engagements rather than self-serve licences. We start with your problem, your data, and the systems you already run.