NeuroRoute NeuroRoute SISL CloudWorx
Login
04NeuroChat/ 10 chapters

The chat they already know. The models, data and cost you control.

NeuroChat is the chat workspace your whole company can use, running on the NeuroRoute gateway. Everything they already expect — documents, projects, artifacts, agents, web search, memory — with 59 models behind one picker, a per-person bill you can actually attribute, and conversations kept under a retention policy you set.

How the routing works
59 modelsone picker, routed per prompt
Per personmetering and chargeback tags
Store nothingor your own retention policy
At a glance
Models in the picker59
Providers behind it4
Measured cost spread48×
DeploymentWhite-label, your domain
04.1

A chat workspace on top of the router

Everything your teams already expect from a chat assistant — documents, files, projects, agents — with the model choice, the bill and the data under your own control.

Every model, one picker

59 models across Anthropic, Vertex, Bedrock and DeepInfra sit behind a single entry in the picker. The router chooses per prompt, so nobody has to learn which model is good at what.

A bill you can actually read

Every message is metered per person and per request, tagged for chargeback, and reported against a fixed baseline. A flat seat subscription cannot tell you which team spent what.

Your retention policy, your key

Conversations are kept only as long as you say — down to storing nothing at all — and content at rest can be encrypted under a key that lives in your own cloud project, which you can revoke.

Your name on it

Shipped white-label: your branding, your own domain. Not a co-branded tab inside somebody else's product.

NeuroChat is an add-on, not a replacement. The gateway, the API and the dashboard are unchanged — NeuroChat is a chat workspace that sits on top of them, licensed separately and enabled per organisation. Teams that only want the API never see it.
04.2

What your people actually see

Two screens carry most of the product: the composer, and the agent marketplace your own teams fill.

NeuroChat, as your people see it. One model picker — NeuroRoute — and the router chooses per prompt behind it, so nobody has to know which of 59 models is the right one, or what it costs. Reproduced here as live markup rather than a screenshot, so it stays legible at any zoom and in either theme.
The in-org agent marketplace: a subject-matter agent is built once, filed under a department, and shared with the organisation — every run still routed, metered and attributed like any other request. Reproduced here as live markup, not a screenshot.
04.3

NeuroChat next to the others

Graded as deployed today. “Others” is the shape of a typical single-vendor seat subscription, and it is strong on most of these rows — the gap NeuroChat is built to close is the last group.

Available todayPartialComing soonNot available
AreaOthersNeuroChat
Models and cost
ModelsOne vendor's own lineup, and only that59 models across Anthropic, Vertex, Bedrock and DeepInfra, behind one picker
Model selectionThe person picks a model by hand, every timeRouted per prompt by task and strategy — a 48× cost spread measured across identical prompts
Cost visibilityNone. A flat per-seat subscription with no per-use breakdownPer-person, per-request metering, chargeback tags, org billing and savings against a fixed baseline
Working with content
Chat with documentsStandardLive — retrieval over your own uploads, proven end to end
File generationUsually via a code sandboxPurpose-built: PDF, DOCX, PPTX and XLSX returned as attachments
Artifacts and canvasStandardSide-by-side artifacts for documents, code and diagrams
Projects and foldersStandardYes
Reading imagesStandardUpload a screenshot, diagram or scan and ask about it
Image generationStandardIn build
Agents and automation
AgentsA large public marketplaceAgent builder plus an in-org marketplace, organised by department
Web searchStandardLive — answers grounded in current sources, with the routing unchanged
Memory across chatsStandardLive — context carries between conversations, under your own retention policy
Connectors and appsVendor-specific app SDKLive over MCP with OAuth 2.1, so any MCP tool connects — not only ones on a vendor's list
VoiceStandardIn build
Scheduled tasksStandardIn build
Identity, data and deployment
Identity and accessVendor accounts, on the vendor's SSO termsYour OIDC. NeuroRoute is the identity provider, with access gated per organisation
Data location and retentionThe vendor's cloud, under the vendor's retention policyYour retention policy, enforced per request — including store-nothing — with at-rest encryption under a key you can hold and revoke
White-labelNot possibleShipped — your branding, on your own domain
Install on desktop and mobileStandardInstalls as an app on desktop and mobile

NeuroChat as deployed, September 2026. "Others" is the shape of a typical single-vendor chat subscription rather than any one product, and it is genuinely strong on the content and automation rows — the difference NeuroChat is built for is model choice, a bill you can attribute, and data that stays under your own policy. Where a row says Coming soon, the capability is in build and is not shipped today. Corrections welcome.

04.4

Identity, metering and guardrails

NeuroChat holds no separate user directory, no separate ledger and no separate policy engine. It is a client of the gateway, so it inherits all three.

NeuroRoute is the identity provider

NeuroChat signs people in through NeuroRoute over OIDC. There is no second user directory to keep in step, and no separate set of credentials to audit.

Access granted per organisation

NeuroChat is a licensed add-on, enabled per organisation with an optional seat limit. Turn it off and the sign-in is refused at the identity layer, not hidden in the interface.

The same metering as the API

Chat traffic is not a separate ledger. It lands in the same usage events, budgets and request logs as everything else, so one invoice and one audit trail cover both.

The same guardrails as the API

PII policy, input and output guardrails, and per-key spend caps apply to a chat message exactly as they do to an API call — because it is the same pipeline underneath.

Nothing here is a second implementation. Read chapter 06 for key custody, encryption at rest and the erasure and export paths — every one of them already covers NeuroChat, because NeuroChat is a client of the same gateway.
04.5

Coming soon

Three capabilities in build. None of them is needed to get value from NeuroChat today, and we would rather list them than let you find out later.

Coming soon
Image generation
Reading an image you upload already works. Generating one needs an image endpoint on the gateway, which is the piece in build.
Coming soon
Voice
Speech-to-text and text-to-speech are wired into the client and waiting on the provider behind them.
Coming soon
Scheduled tasks
Recurring prompts that run on their own and post the result back into the chat.
Why publish this at all? Because a list of everything a product does and nothing it doesn’t is not useful to anyone evaluating it. If one of these is the thing you need, — it moves up the list.