| Provider | Usable now | Implemented | Model-dependent | Partial-compatible | Parameters | Policy-blocked | Vendor-blocked | Unsupported | Total rows |
|---|---|---|---|---|---|---|---|---|---|
| Anthropic | 214 | 99 | 2 | 113 | 88 | 34 | 7 | 39 | 294 |
| Google Gemini | 466 | 463 | 1 | 2 | 394 | 99 | 52 | 79 | 696 |
| Local (on-device) | 185 | 168 | 17 | 0 | 135 | 188 | 6 | 7 | 386 |
| OpenAI | 463 | 452 | 5 | 6 | 461 | 121 | 21 | 77 | 682 |
State semantics
| State | Meaning |
|---|---|
implemented | Wired end-to-end through deeda primitives and probed. |
model-dependent | Works on a documented subset of the provider’s models. |
partial-compatible | Works under a documented compatibility shape. |
policy-blocked | Doc-cited, withheld by deeda policy (e.g. admin surfaces). |
blocked_by_vendor | Vendor has not shipped or access-gates it (sourced + dated). |
unsupported | The provider does not expose this feature. |
How to read a provider page
Each feature family shows two things: the capability rows you can rely on (each linked to the vendor’s primary documentation), and a parameters table — the exact values you can set, their types, defaults, allowed values, and risk level. Parameters withrisk: high typically require
elevated workflow policy (see tool_policy / sandbox).