Leveraging Claude Pro Max subscriptions instead of API keys

[mbenzel]

Looking for workflows that allow us to do hybrid API + subscription activities. This is especially useful to capture the extreme value the subscriptions provide over the raw cost of API usage. For example, the interactive planning while leveraging a high cost model can be done while using a subscription while shorter tasks (reviews, etc) can run autonomously can be handled via agents with the API key. I am unsure what features you would lose in a flow like this and would love to discuss.

[brynary]

We would be happy to support this if it was permitted by Anthropic. However, at this time Anthropic has been ambiguous about what use cases are acceptable for the usage of Claude seat-based subscriptions outside of the Claude Code harness. We're monitoring the situation, and if Anthropic clarifies that it is acceptable to use quad seat subscriptions with open source harnesses like Fabro or OpenCode, then we would be happy to add support.

As an alternative, it is possible to have Fabro run claude by shelling out to it: https://docs.fabro.sh/core-concepts/agents#cli-backend

[mbenzel]

Thanks for the reply - As a company with a smaller budget I am trying to adopt solutions that allow us to hug the value curve and currently the API pricing of the large models is a scary thought when you agentify every step of the SDLC. I was thinking about ways to raise events from the server side that then get caught locally and kicked off from within the subscription context saving on API tokens but still gathering all of the data needed to measure

[brynary]

We now support ACP, which opens up using any agent including Claude Code using it's pre-existing configured credentials! (https://github.com/agentclientprotocol/claude-agent-acp)