Changelog • typesafer

typesafer (development version)

as_tibble() turns a ts_response into a one-row tibble with the same answer columns as system_one_df(), including probs = TRUE.
system_one_df() and as_tibble() gain include_usage to add input_tokens and output_tokens columns.
New ts_default_model() reads the TYPESAFE_DEFAULT_MODEL environment variable, matching the Python SDK, and is now the default model for system_one() and system_one_df(). It falls back to "jev-latest".
system_one(), system_one_df(), and ts_models() gain max_tries and timeout arguments to control retries and the per-attempt timeout for a single call, like the Python SDK’s per-call retry and timeout overrides.
ts_response objects keep the parsed response body in @json, so fields and answer types typesafer doesn’t model yet are still reachable.

system_one() answers now follow the order the questions were asked in, rather than the order the API returns them.

First release of typesafer, an unofficial R client for the TypeSafe System One API.

ts_noul(), ts_choice(), and ts_score() create typed questions as S7 classes, validated before any request is sent.
- Choice criteria accept a bare character vector of options or a named vector or list of descriptions. Options without a description are sent as JSON null. A choice can have at most 255 options.
- Score criteria take 2 to 10 ordered levels and are always sent as a JSON array.
- Noul criteria optionally describe what true and false mean.
- Instructions, options, and levels accept strings or structured lists.

system_one() asks any number of named questions about one state (a string or structured data) in a single request. It returns a ts_response with typed answers, the model that answered, and token usage.
system_one_df() asks the same questions about every row of a data frame, with one request per row, run in parallel with httr2::req_perform_parallel(). It adds a probability column for each noul, a value and _confidence column for each choice and score, and optional _probs list-columns. on_error = "continue" keeps the rows that succeeded when others fail.
Score answers stay on the API’s 0-based scale, with probabilities and level descriptions returned as vectors ordered by level.
ts_models() lists the models available to your account.

Requests are retried on 408, 429, 5xx, and connection failures with exponential backoff and jitter, honoring retry-after-ms and retry-after, matching the official Python SDK’s defaults.
Errors are classed conditions inheriting from typesafer_error:
- typesafer_error_auth
- typesafer_error_validation, with a details tibble of the offending fields
- typesafer_error_rate_limit
- typesafer_error_server
- typesafer_error_http
- typesafer_error_connection
- typesafer_error_input
ts_api_key() reads TYPESAFE_API_KEY and gives a clear error when it’s missing. A rejected key’s error says whether it came from that variable or from the api_key argument.