Where you configure the solution
UI / Proxy, identity provider, MCP server, LLM gateway and agent server. Configure it once, and it keeps production secure, managed and visible.
Solutions
We deliver the product and sample agents. You extend it.
The product securely provides four things: identity, a gateway to models, tools, and controlled agent access. Agents are built on it, and use what it provides. We deliver three sample agents, your team adds its own, and we are happy to help.
What we deliver
Chat, risk management and IVR
Three working Python agents that show how an agent is built and how it joins the platform. Watch the video.
Extend it with your own agents
Agent code is kept simple, on a shared agent library. Identity, guardrails, models, tools and logging come from the platform. We can help you build them.
What you configure
Tenants, users and roles
Create the tenants, then the users in each tenant and the roles they hold. This is the single source of identity for everything else.
Model providers and guardrails
Choose the LLM providers that may be used, and set the guardrails that apply to every model call.
Files, databases, web services and logs
Each is a kind of tool, and you can have several instances of each kind, grouped into knowledge domains. Tools are offered to agents using the MCP protocol.
Constrained access to the tools
An agent is given access to MCP tools, with limits such as select-only on a database. It uses an LLM, through the gateway, to interpret the user's question and to use those tools.
Using all this, you develop agents.
What the product gives you
Secure and easy to manage in production
SAML single sign-on, identity checked on every call, maker and checker approval of every change, and protections against the OWASP Top 10 for the web and for AI.
Visibility into everything
Every call, token and error is logged, and reported per tenant and per user, so you can see who asked what, what it cost and what went wrong.
Agents simply use it
An agent does not build its own sign-in, model access, security checks or logging. It uses the features the product provides to reach its outcome, so agent code stays small.
Tenants, for example
A tenant per department
- HR
- R&D
- Sales
- Finance
- Legal
- Support
HR asks questions of policies and people data. R&D asks questions of research documents. Each team gets its own assistant, answering only from its own information.
A tenant per line of business
- Home Loan
- Credit Card
- Personal
- Business
Home Loans staff never see Credit Card data, and the other way round. Maker and checker approval applies within each line, so every change is accountable to the team that owns it.
Held per tenant
Data
Documents, embeddings and database connections, kept in the tenant's own schemas.
Knowledge and agents
Knowledge domains, and which agents may use which domain.
People and roles
Users, roles and single sign-on, managed per tenant in the identity console.
Approvals and audit
Maker and checker approvals and the change log, per tenant.
Logs and usage
Every call, token and error, reported per tenant and per user.
Any LLM provider
Provider neutral: one governed gateway to the supported providers, with guardrails and rate limits on every call.
Identity everywhere
The tenant and user travel with every call and are validated before any work runs.
Grow tenant by tenant
A new department or business line is a new tenant, not a new installation.
Two ways to run it
As a hosted service, or in your own environment. The tenant model is the same in both.