Skip to content

Coming soon

Planned additions to the core — each ships into the shared foundation, so every connected product benefits from it.

Note

This lists intended direction, not dated commitments.

Connect & extend

How Kiwi reaches beyond its built-in tools — into the MRCL Make modules and into external systems.

  • MCP servers per module — each module (the MRCL Make modules — BrandCheck, GenCanvas, GenScale, GenFlow — and the separate MRCL Ideate) gets its own MCP server that exposes the module's native tools and actions to Kiwi agents. The integration runs both ways: a module's capabilities become tools any Kiwi agent can call, and Kiwi is embedded into the module so its features are driven by those agents. For example, an agent could trigger a BrandCheck evaluation or launch a GenScale batch as part of a larger task. This is the mechanism that turns each product into part of the shared agentic layer.
  • MCP connectors — register external, third-party tools and data sources through the same Model Context Protocol. Once connected, a connector's tools sit alongside the built-in ones and can be granted to any agent — extending what agents can do without changing the platform. Self-serve connector management (adding and configuring connectors yourself) is part of this.
  • Skills — package reusable instructions that teach an agent how to carry out a specific task in a repeatable way, defined on the platform without code. A skill is loaded only when it's relevant, keeping prompts focused, and the same skill can be reused across many agents and applications — so a procedure like a brand-review rubric or a standard report format is written once and shared everywhere.
  • Middleware connection — feed agents additional context data drawn from connected systems — for example, audience or campaign data — so their decisions and recommendations are grounded in current, relevant information rather than the prompt alone.

Orchestrate & reason

How agents handle work that's too complex for a single pass.

  • Planning mode — the agent first breaks a request into an explicit, multi-step plan, then works through it step by step, so complex tasks are handled deliberately rather than in one shot. The plan is visible, so you can see the agent's intended approach — and, over time, steer it — before it executes.
  • Sub-agents — let an agent delegate parts of a task to other, more specialized agents, enabling true multi-agent workflows. Coordination that today lives in the calling application moves into Kiwi itself, with a choice of orchestration architecture to fit the task:
    • Orchestrator–worker — a central orchestrator agent decomposes the task, delegates each subtask to a specialized worker agent, and aggregates the results. Workers don't talk to each other; all coordination flows through the orchestrator, so every decision is easy to follow and trace. The most predictable shape, and the recommended default for most workflows.
    • Swarm — no central coordinator: autonomous peer agents hand off control to one another as the work enters each one's specialty, sharing context as they go. Fewer hops and more resilient (no single point of failure), at the cost of being harder to predict and trace — it needs explicit hand-off and stop conditions so the agents know when to pass work on and when to finish.
    • Custom — compose your own topology — sequential pipelines, parallel fan-out, hierarchies, or a mix — for when neither standard shape fits the task.
  • Deep Research mode — a dedicated mode in which an agent runs an extended, multi-source investigation — searching, reading, and cross-checking across many sources — and returns a structured, cited synthesis rather than a quick answer.

Create & execute

New kinds of output and richer workspaces for producing them.

  • Sandbox mode — a dedicated workspace the agent controls to run code, perform analytics, and build interactive dashboards — going beyond one-off code_interpreter calls to a persistent environment for richer, iterative work.
  • Deck generation — produce ready-to-share presentation decks directly from an agent — for example in PowerPoint (.pptx) format — turning analysis, findings, or campaign concepts into a formatted slide deck without assembling it by hand.
  • 3D rendering — analysis of 3D scenes and objects, plus agent-guided 3D rendering, extending Kiwi's media capabilities into three dimensions.

For what's available today, see Getting started and Use cases; for the products adopting Kiwi, see Powered by Kiwi.