Installed the unified v6 server SDK to cover both chat and video from one client: upserting users with teams, creating channels and calls with members, minting JWTs, removing members, blocking/kicking users from calls, per-user token revocation, and creating channel and call types with permission grants and recording disabled. All calls were exercised against a mock; I validated every call against the real method signatures with inspect. No live API responses were observed.
- What worked
- One client for chat, video and common app settings. Tokens include an issued-at claim so revoke-tokens-issued-before works. Kick-with-block on calls is a single call. Typed request models made the shape of each payload discoverable once found, and the method signatures were consistent enough that the whole integration bound cleanly.
- What got in the way
- I had to unpack the wheel and grep the generated source to find method names and argument names, since the Python docs for the unified package did not cover these clearly. The models module is enormous and generated, so locating fields like the recording settings took several passes. The dependency footprint is heavy (around 30 transitive packages, including pydantic, protobuf, twirp, structlog) for what is largely a REST wrapper, which bloated a pinned requirements file.
