Capability matrix
What each provider integration can and cannot do, and which capabilities NALA cannot determine at all.
Capability varies by provider, and some of it varies by your configuration of that provider rather than by NALA. This page separates the two.
What NALA controls
| Capability | Meaning |
|---|---|
| Launch | NALA can start the provider in a managed pane |
| Structured session | The provider can be driven as a session with events, not just a terminal |
| Native TUI | The provider's own terminal interface is used directly |
| Resume | A session can be reattached after detaching |
| Cancel | An in-flight turn can be cancelled |
| Task assignment | The provider can be given a durable task |
| Artifact access | The provider can read and write artifacts |
| A2A send / receive | The provider can participate in agent-to-agent messaging |
What the provider controls
| Capability | Determined by |
|---|---|
| Model selection | Which models your plan exposes |
| Effort / reasoning depth | Whether that provider exposes the concept at all |
| Rate limits | The provider's service |
| Context window | The model |
IMPORTANT
NALA cannot enumerate the models your account is entitled to. It can pass a model through; it cannot tell you in advance whether your plan permits it.
The three states, again
Reading this matrix without holding these apart will mislead you:
| State | Who knows |
|---|---|
| Detected | NALA |
| Authenticated | The provider |
| Entitled | The provider's billing |
A provider row in the registry means NALA knows how to launch it. Nothing more.
Capability negotiation
For external harnesses driven over the Agent Client Protocol, session capabilities are negotiated at handshake and reported on the launch card as they were actually negotiated — not as a static list.
This means the same harness can present different capabilities on different machines or versions. Trust the launch card over any documentation, including this page.
Billing class
Every route carries a billing class: whose money is spent and under which plan. Two provider integrations that look identical in this matrix can bill very differently — one against a subscription, one against an API key.
See providers, models, and routes.
What is not claimed
- No provider's quota, plan tier, or remaining spend is shown.
- No ranking of providers or models is published. Which one is "best" depends on the task and the repository.
- Capability tables are not a promise that the provider is installed on your machine or working today.
Verifying on your own machine
nala doctor
Reports what NALA actually resolved: found or not, resolved path, detected version. That is the honest boundary of NALA's knowledge.