deeplinq is built from four primitives — Connectors, Agents, Apps, and Channels — and one published SDK. Our own agents and first-party apps run on the exact same public wire as anyone else's. No back doors. No internal shortcut. The neutrality you're sold is enforced by the architecture, not promised in a contract.
Every workload on deeplinq resolves to the same four building blocks. Configure them once per tenant, reuse them everywhere. The build view stays the same whether you ship a single FAQ agent or a full operational app.
Credentialed adapters to the systems you already run — CRM, messaging, search, vector stores, model providers, MCP tools. Configure each once per tenant. Every agent and app reuses it. Your credentials stay in your perimeter; the platform meets your stack where it is.
A headless AI brain defined by a versioned manifest. Run a stateless RAG or classifier in manifest mode, or a long-running, human-in-the-loop workflow in programmatic mode with ReAct and sub-agents. Host the runtime with us, or point to your own — both answer on the same wire.
First-class products with their own UI, business logic, and data, installed per tenant. The agent owns the brain and its safety; the app owns operational policy — approvals, SLAs, escalation, rate limits, the screens your team actually uses. Studio Chat, CRM Suite, and partner apps are all just apps.
Thin transport adapters bound one-to-one to an app. WhatsApp, SMS, email, web, CLI. The work doesn't change with the surface — the same governed app answers wherever your users already are.
@deeplinq-org/sdk is one unified TypeScript SDK for both Agents and Apps. There is no internal edition. The SDK our own teams ship products with is the SDK you install. What we can build, you can build — against the same documented surface.
deeplinq's own agent runtime, core tools, and first-party apps run on the exact same public wire as partners — OAuth client credentials, REST, SSE, HMAC-signed webhooks. No privileged bypass. We don't give ourselves a path we won't give you.
deeplinq platform
Public wire
OAuth · REST · SSE · HMAC
Same wire. No privileged bypass.
deeplinq first-party apps
Studio Chat · CRM Suite
Partner apps
Partner app · Partner app
Whether inference runs in our standard runtime or a partner-hosted one, every call goes through a single governed endpoint — metered, audited, policy-enforced. Manifests declare a preferred model and the capabilities a task needs; the gateway resolves across providers at invoke time. Cloud APIs for non-sensitive work — OpenAI, Anthropic, Mistral, Google. Open-weights, self-hosted, for on-premise or air-gapped — Llama, Qwen, Mistral, Gemma, Falcon. Partners never bring their own model credentials. The choice of model belongs to the institution, and switching it never touches your code.
The four primitives are the build view — how you compose what ships. Underneath, the same platform runs as four layers at runtime: Connector Hub, RAG Engine, LLM Router, Agent Orchestrator, with RBAC and an append-only audit trail wrapping every one. Build view to compose, runtime view to govern.
See the runtime architecture →Connect your systems, ship governed agents and apps, and route every model call through one auditable gateway — inside the environment you already defend. We'll walk your team through the SDK and the wire.
Book a platform briefing