Agentic AI Comparison:
Dust AI vs Kohala

Dust AI: 8/10

Evidence confidence: high

Dust has a substantial documented end-to-end workflow for its stated purpose: users can create custom agents combining models, instructions, tools, and company knowledge; select among models from OpenAI, Anthropic, Google, Mistral, and open-source providers; synchronize selected enterprise data sources; and use agents to answer questions, draft documents, analyze data, and act in connected systems. The official product description also documents reusable skills, agent memory, shared workspaces, automations, and multi-agent collaboration. Controls are documented around synchronized data and permissions: administrators choose what data is synchronized, and the documentation specifies limits for sources such as Slack, Google Drive, GitHub, Confluence, and Intercom. The score is below 9 because the available evidence describes a broad platform but does not independently establish the completeness or effectiveness of every advertised workflow, and some advanced capabilities are described at a product level rather than through detailed implementation evidence.

Kohala: 6/10

Evidence confidence: medium

Kohala has a concrete but narrower documented workflow: agents, scripts, timers, and human-approval steps can be orchestrated into durable workflows that teams can run, monitor, and replay. MCP support adds a documented mechanism for connecting external tools and exposing Kohala agents to MCP clients. This supports a meaningful automation and approval use case, but the available primary-source evidence does not establish the breadth of workflow primitives, agent configuration, data grounding, permissions, observability, failure handling, or domain-specific capabilities needed for a higher score. The score reflects a useful documented core rather than a claim about measured execution quality.

Dust AI: 8/10

Evidence confidence: high

Dust is presented as an active, accessible product with a free Business option requiring no credit card, supporting up to five users, three connectors, and five Spaces; paid Pro and Max seats are also documented. The documentation provides an introduction, rollout guidance, agent creation material, and managed connection workflows, while the product supports business-user agent creation without requiring every user to build an application from scratch. Adoption still requires workspace administration, data synchronization choices, model and connector configuration, and potentially enterprise implementation, so the score is not 9 or 10.

Kohala: 6/10

Evidence confidence: low

The official Kohala source documents a Business Account priced at $30/month and describes a team-oriented product that can run and monitor workflows. That establishes a current-looking access path, but the available evidence does not verify whether signup is self-service, whether a trial or free tier exists, how agents and scripts are configured, what hosting or model setup is required, or whether MCP connections require administrator or developer work. Because the intended user can be identified but the practical onboarding path is incompletely documented, the score is conservative.

Dust AI: 8/10

Evidence confidence: high

Dust provides unusually clear entry pricing for the documented scope: Business Free is listed at €0 with 500 lifetime credits, Pro at €24 per seat per month with 8,000 credits per seat per month, and Max at €120 per seat per month with 40,000 credits per seat per month. The pricing page also describes model access, custom agents, scheduled and triggered multi-agent workflows, connectors, MCP, and collaboration features. Enterprise pricing is not presented as a simple public amount, and usage is credit-based, so buyers must account for seats, credits, model consumption, and potentially enterprise terms; the documentation also says programmatic usage is tracked and billed separately and that enterprise users can use pay-as-you-go billing. The score recognizes strong published clarity without assuming that the listed subscription includes unlimited model or compute usage.

Kohala: 5/10

Evidence confidence: low

Kohala's official Business Account page gives a visible price of $30/month and identifies a useful scope: durable workflows, agents, scripts, timers, human approvals, monitoring, replay, and MCP connectivity. However, the available source does not clarify whether the $30 price covers users, workflow runs, agent execution, model usage, storage, or hosting; it also does not establish enterprise pricing, limits, trials, or additional usage charges. Unknown price details are not treated as zero cost, but they materially limit budgeting confidence, producing a mid-range score rather than a low score.

Dust AI: 9/10

Evidence confidence: high

Dust documents broad user-facing integration options. Its pricing page lists Slack, Notion, GitHub, Drive, more than 20 additional connections, and any tool through MCP. The official integrations page lists tools including Airtable, Amplitude, Costory, NetSuite, and Power BI, while the product documentation describes managed connections such as Google Drive, Notion, Slack, and Confluence. Dust also documents API access and a Chrome extension for bringing agents into existing workflows. The score is below 10 because the available evidence verifies breadth and extensibility but does not establish that every listed connector supports every action or that all integrations are available on every plan.

Kohala: 5/10

Evidence confidence: medium

Kohala has two verified integration mechanisms in the available primary source: agents can connect to external tools over MCP, and Kohala agents can be exposed as secure MCP servers callable by clients such as Claude and Cursor. This is meaningful extensibility, but it is not evidence of a broad native connector catalog, direct integrations for specific business systems, import/export coverage, or a public API beyond MCP. Internal workflow components such as scripts and timers are capabilities, not automatically separate external integrations. The score therefore reflects a useful configurable connection layer with limited verified breadth.