Cana

Cana is an agentic platform from Colate Ltd for building, governing and deploying AI agents, including what it calls Super Intelligent Agents (SIAs) that own whole outcomes within a function. It spans a desktop app, web, iOS, Android and a Chrome extension, and early access is currently open through a waitlist.

What It Is

Cana starts as a chat: users pick a mode (Ask or Workbench, where the agent gets its own computer) and a permission level that controls how freely the agent may act, from asking before anything runs to running unchecked. From there the platform offers Cana Studio for pages and small apps, Cana Code for agentic engineering (agents, editor, browser preview, terminal and Kubernetes tools), MCP Hub and Gateway for connecting tools, Cana Twins for embedding assistants in a product, plus Relays, Meetings, Boards, Skills & Knowledge, Artifacts and Mods.

Walkie-Talkie and multi-agent work

Walkie-Talkie is a live channel where people and agents such as Claude Code, Codex, Cana chats and bots talk to each other. Boards give people and agents shared task lists, and Relays are chats that keep working on a schedule or on events after the user leaves.

Governance and deployment

Cana states that every agent gets an identity, permission scope, budget and audit trail before acting. It lists a policy engine, evaluations, shared memory, a knowledge graph and a model router, with models from Anthropic, OpenAI, Google Gemini, Kimi, xAI and any OpenAI-compatible or private model. Deployment options listed include Cana Cloud, a customer VPC on AWS, Google Cloud, Azure or Oracle, on-prem, and air-gapped. Enterprise pages mention SSO, SCIM, RBAC and SIEM-exportable traces.

For developers

The features page describes TypeScript and Python SDKs, a CLI and a REST API for creating agents, attaching MCP tools, setting policy and running them. Six SIAs are announced for Consulting, Telecom, Engineering, AIOps, Customer Experience and Custom functions.