Skip to content
Products/Operate layer/ChatLite AI
Operate layerGenerally available

ChatLite AI

Every leading model. One governed workspace.

ChatLite is enterprise-first. Deploy the whole platform inside your own cloud or data centre, connect your internal systems, live data feeds and custom agents, and give every department one place to work with OpenAI, Anthropic, Google, Mistral, DeepSeek and xAI models. Guardrails, SSO and audit logs are on from the first prompt, not bolted on after the pilot.

Six model providers
OpenAI, Anthropic, Google, Mistral, DeepSeek, xAI
Fully self-hostable
Your cloud or your data centre
Around 30 file formats
PDF, DOCX, XLSX, CSV and source code
Policy checkSELF-HOSTED
Prompt · marketing workspace
Draft a launch email referencing the customer list in the attached file.
Prompt injectionCLEAR
PII detected in attachment24 EMAILS REDACTED
Content policyWITHIN POLICY
Model output screenALLOWED
outcome recorded in the audit logALLOWED WITH REDACTIONS
Model providers and sources it routes to
OpenAIAnthropic ClaudeGoogle GeminiMistralDeepSeekxAI GrokCohereDALL·EImagenFluxStable DiffusionIdeogramRecraftWolfram AlphaGoogle DriveOneDrive
Platform

Minimal interface. Serious capability underneath.

The chat surface stays light so people actually use it. The platform beneath gives you the model choice, the security posture and the control an enterprise rollout needs.

Every model, one workspace

Switch between reasoning and multimodal models from six providers without changing tools or losing the thread you are working in.

Governed by default

Central access control, role-based permissions and admin controls so adoption does not cost you oversight.

Web search, grounded

Answers tied to live sources with citations your team can verify before anything is forwarded or published.

Files and data

Summarise documents, analyse spreadsheets and read around thirty formats, or pull files straight from Google Drive and OneDrive.

Code and execution

Sandboxed Python for real analysis, generated charts and Wolfram Alpha for exact computation, orchestrated without leaving the chat.

Image generation

On-brand visuals, diagrams and concepts from seven image providers, in the same workspace as the text work.

Team workspaces

Shared prompts, threads and presets so a good approach spreads instead of being reinvented in every department.

Custom agents and feeds

Internal systems, live data feeds and your own agents brought onto the same governed plane as chat and research.

Deep research

Open questions become source-backed reports.

ChatLite plans the research path before it searches, gathers live web and internal context, weighs the sources against each other and returns a structured brief. Different models can be assigned to the planner, researcher, verifier and writer roles, so the checking is not done by the model that wrote the claim.

Step 01
Plans before it searches

The planner drafts a search path and states what evidence would answer the question, so the run can be reviewed rather than trusted.

Planner role, assignable per model
Step 02
Reads live and internal sources

Web sources sit alongside your uploaded documents and connected systems, in one pass rather than two disconnected ones.

Press, filings and analyst notes in the same brief
Step 03
Weighs the claims

Conflicting sources are compared and source quality is shown next to every claim, so a weak citation is visible rather than averaged away.

Verifier role, separate model
Step 04
Returns something reviewable

Findings, risks and recommended next steps in a structured brief your team can share, challenge or build on.

Exportable, with citations attached
Guardrails

Safe by default, because the policy runs before the model does.

ChatLite enforces your rules on every prompt and every response. The same ZNYX policy engine that governs your custom applications governs the workspace your staff type into.

PII and secrets
Detected and redacted before a prompt reaches a provider, with the redaction recorded rather than silently applied.
Jailbreak defence
Prompt-injection attempts and jailbreak patterns are blocked at the boundary, including instructions hidden inside uploaded files.
Policy per team
Content policies tuned by department, so legal and marketing are not forced to share one threshold.
Full audit trail
Every decision, redaction and block is logged with the policy that caused it, which is what an auditor asks to see.
For teams

A practical assistant for every department.

From internal knowledge work to customer-facing communication, with usage consistent enough that an administrator can explain it.

Internal knowledge discovery

Find the answer that already exists inside your own documents instead of asking the person who wrote them four years ago.

Every department
Research and competitive analysis

Deep research runs produce cited briefs on markets, competitors and regulation, with the sources open to inspection.

Strategy and product
Customer support drafting

Consistent, on-policy responses drafted from your real help content, with the agent still deciding what to send.

Service teams
Data analysis and reporting

Spreadsheets analysed in a sandbox, charts generated, and the working shown rather than asserted.

Finance and operations
Code, docs and review help

Read source files, draft documentation and review changes in the same workspace the rest of the business uses.

Engineering
Deployment

Self-host it. Customise it. Connect your agents.

Chat, deep research, image generation, web search, your internal data feeds and your own agents all sit on one governed plane. Where that plane runs is your decision, not ours.

Where it runs
Your environment

Deploy the entire platform in your own cloud or data centre so prompts, files and outputs never leave it. This is what makes data residency answerable.

Identity
SSO, MFA and RBAC

OAuth-based single sign-on, TOTP multi-factor with backup codes, device and session control, and role-based access throughout.

Integration
Your systems and feeds

Custom integrations and live data feeds connect internal databases and knowledge sources, so answers are grounded in your own data.

Training data
Never used for training

Prompts, files and outputs are processed to provide the service and are not used to train foundation models unless you explicitly opt in.

Common questions

What administrators ask first.

ChatLite runs for individuals on a free plan and for organisations under central administration. If your question is not here, ask it directly and you will get a straight answer.

Talk to us
Which models does ChatLite support?

Leading models from OpenAI, Anthropic, Google, Mistral, DeepSeek and xAI, including reasoning and multimodal models, from one workspace and without losing context when you switch.

Can it be self-hosted?

Yes. The whole platform can be deployed inside your own cloud or data centre, which is how strict residency and compliance requirements get met rather than argued about.

Do you use our data to train models?

No. Prompts, files and outputs are processed only to provide the service and are never used to train third-party foundation models unless you explicitly opt in.

Does it support SSO and MFA?

Yes. Organisations sign in through OAuth-based single sign-on, and any account can enable TOTP multi-factor authentication with backup codes.

Can we connect our own agents?

Yes. Custom agents and tools are orchestrated alongside chat, deep research, web search and image generation on one governed plane.

What is deep research mode?

A run that plans a search path, gathers live web and uploaded sources, compares claims and returns a cited brief. Planner, researcher, verifier and writer roles can each use a different model.

Bring secure AI into everyday work.

Start with one department and a policy you already have written down. We will run the pilot in your environment, with your SSO, and show you the audit trail before you widen access.

Talk to us