Run the work
Goals Playbooks AI coding agents Ask KadmoEquip the agents
Claude Code skills Skill packs MCP servers IntegrationsKeep control
Self-hosted agents Guardrails Models and keys ReportingClaude can remain part of your team’s work without becoming its only model family. Kadmo separates the work platform from the supported providers that serve it, giving your organization a place to manage provider connections, model availability and data rules.
Choose which services the team may use, under whose credentials and within which permitted regions. Keep those decisions explicit as your work and providers change.
Three decisions sit behind a model request. A company makes the model. A service hosts its inference. A work platform supplies the context, instructions and process around the call.
Accessing Claude through another cloud changes the hosting relationship while retaining Anthropic’s model family. Kadmo adds the ability to use supported models from other makers within the same work platform. Decide which kind of choice your organization needs, then assess the actual route a request will take.
Anthropic models · Native coding CLI
Run Claude Code directly with Anthropic models as a Kadmo agent. Use the native coding CLI within the roles and processes your team assigns, with the job’s progress and results available in Kadmo. This is the straightforward option when you want Claude Code connected to Anthropic.
Anthropic models · Routed coding CLI
Connect Claude Code to Anthropic through Claude Code Router. Choose this arrangement when your agent setup uses the router to handle its model connection while keeping Claude Code as the coding CLI. Kadmo supports this alongside the direct Claude Code option.
An admin chooses the account’s available models. Model entries show where inference is served, so the team can distinguish the location of the Kadmo app from the destination of a model request.
With Business and up, regional policy can narrow the permitted destinations for the account and its workspaces. A call outside the applicable regions is refused with a reason. An EU-only policy can use only enabled models served there; if none meet the policy, it leaves no eligible route. The policy enforces the destination you permit.
Kadmo-supplied model access uses the account’s Kadmo credit. With Enterprise, you can connect your own provider keys or use supported gateway and provider arrangements. Calls using your own credentials belong to that provider billing relationship.
The account’s provider supplies the default, while an agent’s configured app or a run’s explicit choice can determine its own connection. Make those assignments deliberate so the team understands which provider serves a job. The coding CLI and the source of model access are separate parts of the setup.
If your chosen provider key cannot serve a request, the request is refused with the reason and the action needed. Kadmo does not silently switch that failed own-key request to its prepaid model credit.
That behavior keeps the supplier and payer you selected explicit during failure. An administrator can repair the connection or intentionally choose another supported arrangement.
Content rules give admins on Business and up three responses to matched chat content: audit it, mask it or block the message. For example, a rule can replace a detected email address before a message is sent, or refuse a message containing a detected API key. The account can add patterns for its own identifiers.
These rules cover outbound chat messages in the supported Kadmo chat surfaces. They do not inspect attachments, tool results, model answers, agent-job prompts or a member’s external-client calls. Use their actual coverage when deciding which work and data belong in a given route.
Kadmo operates the workspace and coordination service. The selected model service processes the request it receives. Your agent environment and any connected tools are additional parts of the data flow. App hosting, code execution and model inference therefore need separate answers in a procurement review.
Review the processors, training treatment, retention and deletion terms for the services in your chosen arrangement. Kadmo’s privacy policy names its processing relationships; Enterprise arrangements include the data processing agreement and contractual review your procurement requires. A model or region setting supports those decisions and should be assessed alongside the applicable agreement.
If a single Anthropic service covers your workload and its direct agreement meets your requirements, using it directly may be the shorter path. Kadmo fits an organization that also wants its work platform and operating rules to accommodate other supported providers.
For choosing models by quality and cost within a workflow, see the OpenAI alternative. For project questions and reusable knowledge, see the Claude alternative.
No. Retain Claude for suitable work and add another supported model where your organization approves it. The benefit is keeping the provider decision available as requirements change.
No. An agent can run on your machine while sending model requests to a remote inference service. Assess its configured model route separately from where the repository and coding process run.
Yes. A workspace can narrow the account’s permitted regions. It cannot make a region available that the account disallows, and its eligible models must also be enabled and included in the plan.
Admins manage the provider connections, approved models and applicable guardrails. Members can inspect the available settings and use the models and tools their account permits.
Choose a representative piece of work and agree its permitted destination, credential owner, data handling and approval conditions. Run it through the proposed arrangement and review the evidence with the people responsible for your platform and data.