Everything AgentGate does
Connect your systems, govern how they are used and control the models, in a single platform.
Connect
Your systems become business tools that assistants can use.
AI assistant connection
ChatGPT, Claude, OpenCode and any MCP-compatible client, with your organization's sign-in (SSO and MFA).
Connector in your network
Installed with Docker. It only opens an outbound, encrypted, certificate-based (mTLS) connection; it opens no inbound ports.
Direct mode (optional)
The platform connects on its own to reachable databases and APIs, with nothing to install. The credential is stored encrypted.
Databases
PostgreSQL, MySQL, MariaDB, SQL Server and Oracle, with predefined, parameterized queries.
APIs and other servers
REST, HTTP and OpenAPI APIs, and also tools from other MCP servers.
MCP gateway
Brings together, in a single point, the MCP servers that have no security of their own and cannot be exposed. They stay behind AgentGate, which adds sign-in, permissions per tool and a record of every call.
Custom tools
Multi-step tools in one transaction, tools composed of several systems, or tools that ask a model. AI-generated drafts from a description.
Modules that build tools
A module groups the tools, skills and commands for a task under a single permission. You describe the requirement in words or documents, the AI reads your connectors and proposes which tools to build, and you choose which ones. They are implemented and tested, and nothing is published until a person reviews it.
Managed from the assistant itself
Administrators can create tools and groups from the assistant or through the API.
Instructions for assistants
Ready-made commands and instructions the assistant reads from the server, and notices when tools change.
Govern
Control over who does what, and a record of everything that happens.
Read-only by default
Every connector rejects writes unless you enable them.
Change confirmation
Every tool that writes asks for confirmation before it runs.
Denied by default
Each tool is assigned to roles or groups of the organization.
Approved tools
You approve specific tools; if someone changes what they do, they stop running.
Signed catalog
The connector only accepts signed, current catalogs.
Limits per tool
Row caps per query and per write, and minimums and maximums per field.
Natural-language rules
A condition written in words, such as “the discount cannot exceed 20%”, which a model checks before running. Depends on the plan.
Audit trail
Who called, from which client and with what result, with an executions console.
Everything in one place
You govern both tools and models: you know which model each person uses and which MCP servers they use, without stitching together logs from several systems.
Less Shadow AI
When there is no official path, people use personal AI accounts with company data. AgentGate gives them approved access to models and tools, with their own key, personal data masked and a record of what is used, and you see what each person does.
No duplicates
If the agent retries, the operation is not repeated. Quotas per organization.
Protected secrets
Passwords are stored encrypted and by reference, and certificates rotate without stopping the service.
Control the models
The AI model gateway: one access point with budgets and rules.
LLM gateway
Centralizes calls to AI models across the whole organization: one place to see who uses which model, how much they spend and what data leaves.
A single access point to models
Compatible with OpenAI and Anthropic, for the tools your team already uses.
Several providers
OpenAI and compatible ones (such as Azure, vLLM or Ollama), Gemini and Anthropic.
Budgets
Per organization, group, key, person and session. With no balance, the model is not called.
Virtual keys
Each key has one holder and is revoked when the person leaves the group.
Every session in full
The model, the tools and the cost of a session, in a single view.
Security steps
Before and after the model and the tools: mask personal data and apply rules.
Privacy
Content is processed in your network; only metadata reaches the platform. Content analysis is optional.
Works with the AI you already use and the systems you already have
Assistants and agents
Claude, ChatGPT, OpenCode, Hermes, OpenClaw and any MCP-compatible client.
Systems
PostgreSQL, MySQL, MariaDB, SQL Server, Oracle, REST and OpenAPI APIs, SOAP and MQTT.