Roadmap#

Maintained by @hacker-no-ice, currently the sole maintainer. Themes are listed in the order they are planned. They are directions, not dated commitments, and they change as the project learns from people using it.

01 · INDIVIDUALSEasy & safe locallyLess setup friction.
Clear costs. Safe defaults.
02 · WORKLOADSGovern model callsAdvanced models.
Same policy and ledger.
03 · DEPLOYMENTCloud agents & teamsLinux deployment.
Shared budgets and roles.

Planned order, not release dates. Each priority builds on the same spending boundaries.

1. Easy and safe for individuals#

The first priority is making Hubu easy and safe for one person running it on their own machine.

Easier to use

A shorter path from installation to the first governed request.
Clearer setup, diagnostics, and error messages, so problems explain themselves.
A simpler approval flow in the agent client.

Safer to use

Safe defaults: conservative budgets, with approval required unless you choose otherwise.
Clear previews of what a request will cost before you approve it.
Continued hardening of how keys and approval authority are handled.

2. Governed model calls#

Today Hubu governs image generation. The next workload is calls to advanced AI models.

For example, you use a low-cost agent subscription but want the agent to call a more capable model for some tasks. Hubu would govern those calls the same way it governs image generation: policy and approval, funds reserved before the call, and the actual cost settled in the same ledger. The agent still never holds the API key.

Gongbu integrations for calling models directly through a provider's API, or through an API gateway you already use.
No change to the architecture: Hubu governs, Gongbu executes, and agents use the same MCP surface.

3. A deployed Hubu for cloud agents and small teams#

Hubu runs on your own machine today, and only that machine can reach it. Persistent agents such as ChatGPT dots, Grok Bot, and Muse run in their own cloud computers, where they cannot reach a stack on your laptop, and teammates cannot share it either.

A deployed version of Hubu that these persistent cloud agents can connect to, so they get the same per-agent identity, policy, approval, budgets, and ledger.
Linux support, since a deployed Hubu runs on servers rather than a Mac.
Human approvals still come from you or the people you authorize, never from the agent's environment.
One deployment per team, so a few people can share budgets and approvals, with roles for who can approve spending or change budgets.

Not the focus right now#

These are reasonable requests, but they are not planned for the near term. If there is strong demand from people using Hubu, the focus can shift, so tell us what you need.

Broad provider coverage beyond the image and model workloads above.
Platforms beyond macOS and the Linux deployment above.
Guided setup for more local agent clients. Other MCP clients already work with manual configuration.
A production payment rail.
Multi-tenant hosting, where many separate organizations share one service.

Get involved#

Hubu welcomes contributors. If this direction is useful to you, open an issue to discuss an idea or say what you would like to work on, or report a bug.