The system of record behind the whole platform. It runs your agents, executes tools over MCP, and owns conversations, memory and knowledge. The chat app renders and the gateway streams — everything they do flows through here, and through the same REST API and MCP server you can call yourself.
{
"method": "tools/call",
"params": {
"name": "search_knowledge",
"agent": "support-bot",
"arguments": { "query": "refund policy" }
}
}
Found 3 relevant chunks · cosine ≥ 0.81
Injecting context into the model▋
One platform · any model — bring your own keys
The chat app renders and the gateway streams — but agents, tools, memory and knowledge all live in the Core Service. It's also a product on its own: anything the stack does, you can do directly over the API.
The gateway is just one client — point any MCP client or HTTP request straight at the Core Service.
Spec-compliant JSON-RPC 2.0. Connect any MCP-capable client to per-agent tools.
POST
/api/v1/mcp/handle/{agent}
Agents, conversations, messages, memory, feedback and usage — all under one versioned namespace.
GET
· POST /api/v1/*
A Passport provider with personal access tokens and role-based permissions — the gateway is just an OAuth client.
Bearer
· Laravel Passport
Not just another chat box. A full backend for agents, tools, knowledge, and memory — with the admin controls to run it in production.
Spin up specialized assistants, each with its own model, system prompt, temperature, tool set, and knowledge base. Lock the model or agent selector per use case, set welcome and refusal messages, and toggle memory — no code required.
A spec-compliant JSON-RPC 2.0 MCP server with auto-discovered tools, per-agent scoping, and 60s session caching at the gateway.
Register external APIs with a clean Service → Tool → Endpoint model. Bulk-import from Swagger, attach encrypted auth, and call them from any agent.
Upload PDFs, docs and FAQs per agent. Automatic chunking, embeddings and cosine search inject only the most relevant context into each answer.
Three tiers in one: short-term context windows, async summarization for long threads, and long-term per-user memories the assistant recalls automatically.
Ship a white-label assistant into any site or app with iframe auth and guest sessions. Lock it to one agent, match your brand, and track anonymous visitors — streaming responses, tool calls and rich UI included.
OAuth2 (Passport) SSO, granular roles and permissions, personal API tokens, and full request audit logs.
Per-user token budgets, monthly limits, image-generation caps, and a live dashboard of consumption across every model.
No glue code. Wire up a capability in the dashboard and your agents can use it on the next message.
Add any external API as a service, define its endpoints and methods, and attach encrypted authentication.
Bind tools to an agent with priority and per-tool instructions. Add knowledge files for grounded answers.
The model discovers tools over MCP and calls them mid-conversation — results stream straight back to the user.
Open the dashboard to create your first agent, register a tool, and watch it call your APIs in real time.