* fix(kosong): make OpenAI-compatible thinking work without reasoning_key
Reasoning field names (reasoning_content / reasoning_details / reasoning)
are protocol facts, not user preferences. Treating reasoning_key as a
required user-set field meant any path that didn't go through the catalog
— hand-written config.toml in particular — silently lost thinking content
and broke strict gateways like DeepSeek.
Demote reasoning_key to an internal protocol constant with an explicit
override:
- Inbound (stream + non-stream): scan reasoning_content,
reasoning_details, reasoning in order; first string value wins. An
explicit reasoning_key restricts the scan to that one field.
- Outbound: serialize ThinkPart back as reasoning_content by default.
An explicit reasoning_key writes to that field instead.
- reasoning_effort auto-injection no longer requires reasoning_key;
presence of ThinkPart in history is enough.
Catalog plumbing is unchanged — explicit values from the catalog still
win, the default just stops being undefined.
Manually verified end-to-end against the real DeepSeek API with a
hand-written config.toml that does not set reasoning_key: thinking
content renders, no 400, multi-turn conversations work.
* fix(kosong): normalize blank reasoning_key to unset
ModelAliasSchema accepts `reasoning_key = ""` (z.string().optional()).
A blank value used to disable the default field scan and route both
inbound reads and outbound writes through an empty property name.
Trim and treat empty as undefined at the provider boundary so the
default protocol behavior applies.
* fix(kosong): preserve caller-pinned reasoning_effort during auto-inject
When the history contains ThinkPart, generate() injects
reasoning_effort='medium' and then assigns it onto createParams,
which used to silently overwrite a value the caller set via
withGenerationKwargs({ reasoning_effort: 'high' }). Skip auto-inject
when an explicit reasoning_effort already lives in kwargs.
* fix(catalog): preserve reasoning fields
* fix(catalog): treat interleaved=true as reasoning_content
models.dev documents `interleaved` as `boolean | { field }`, where the
bare boolean means "general support" without an explicit field name.
The previous branch returned undefined for `true`, leaving openai-compat
gateways that publish `interleaved: true` without a round-tripped
reasoning field. Map `true` to the default `reasoning_content` so those
models still surface and replay thinking content.
* fix(catalog): preserve interleaved field in built-in catalog snapshot
`update-catalog.mjs` drives the bundled catalog that ships with release
builds and is the default source for `/connect`. The allowlist dropped
`interleaved`, so even after the runtime learned to read the field, the
default offline path never sees it — reasoning round-tripping silently
stayed off for openai-compat models in release builds. Keep
`interleaved` so the bundled snapshot carries the same metadata as the
live models.dev catalog.
---------
Co-authored-by: 7Sageer <7sageer@djwcb.cn>
* feat: add /connect command with models.dev catalog support
Add a /connect slash command that configures a provider and model
from a models.dev-style catalog. Users no longer need to hand-write
model metadata (context window, output limit, capabilities).
Architecture (3 layers):
- kosong: pure data layer — Catalog schema, inferWireType,
catalogModelToCapability
- node-sdk: IO + config write — fetchCatalog, applyCatalogProvider,
catalogModelToAlias
- app: TUI flow — /connect command, provider/model selection,
credential input, config persistence
UI improvements in this PR:
- ChoicePickerComponent: add searchable (fuzzy filter + search bar)
- ModelSelectorComponent: add searchable (same)
- Extract reusable paging.ts for list pagination
Changesets included for kosong, kimi-code-sdk, and kimi-code.
* feat: bundle pruned models.dev catalog for offline /connect
When the network is unavailable, /connect now falls back to a built-in
snapshot of the models.dev catalog.
- `scripts/update-catalog.mjs`: fetches models.dev/api.json, strips
unnecessary fields, and writes `src/built-in-catalog.ts` with the
JSON string as a TS constant.
- `loadBuiltInCatalog(text?)` in node-sdk: parses the JSON string safely;
returns undefined on any failure.
- `handleConnectCommand`: on fetch failure, shows an informative offline
message and tries the built-in snapshot.
- The snapshot file is a placeholder (`undefined`) in source control;
`update-catalog.mjs` populates it before release builds so the actual
catalog is inlined into the bundle by rolldown.
* refactor(tui): share search and pagination across list pickers
ChoicePicker and ModelSelector each carried their own copy of the cursor +
fuzzy-search + pagination state machine. Extract it into a reusable
SearchableList so both pickers share one implementation; behavior is unchanged.
* fix(tui): filter unsupported catalog providers
* docs(tui): clarify /connect stale-alias cleanup depends on removeProvider
* fix(kimi-code): inject built-in catalog at release time
* feat(tui): hint at /login and /connect when /model has no models
Replace the bare "No models configured." error with a notice that
points users to /login for Kimi and /connect for other providers.
* fix(tui): tighten /connect error reporting for edge cases
Two silent-failure cases in /connect could leave users without any
feedback to act on:
- Reject `--url` when its value is missing (e.g. `/connect --url` or
`/connect --url=`). Previously the argument parser silently fell
back to the default catalog, so a malformed flag still appeared to
succeed but with the wrong source.
- Show an explicit error when the resolved catalog yields no providers
with supported wire types. Previously the picker resolved with no
selection and the command returned without any UI feedback.
* chore(tui): restore slash invalid intent type
* fix(tui): support /logout for /connect-configured providers
After /connect writes a non-managed provider (e.g. openai), /logout
fell through to "Nothing to logout." because the handler only matched
the managed default and isOpenPlatformId branches, leaving users no
in-app way to drop the API key and model aliases they just configured.
Collapse the OpenPlatform branch into a generic "provider is present
in config" check so any non-managed provider in config — OpenPlatform
OAuth targets and /connect catalog providers — goes through the same
removeProvider path.
* fix(tui): reject --url values that are not http(s) URLs
`resolveConnectCatalogRequest` previously matched any non-space token
after `--url` as the URL, so `/connect --url --refresh` parsed
`--refresh` as the value and bypassed the missing-value error path.
Bare non-URL tokens (`/connect --url not-a-url`) and non-http(s)
schemes were also silently accepted.
Constrain the captured value to `https?://...` so flag-like and
non-URL tokens fall through to the existing `URL_FLAG_PRESENT_RE`
check and surface a clear error.
* ci(native): scope built-in catalog generation to signed macOS jobs
The catalog-generation step ran whenever `inputs.sign-macos` was true,
including Linux and Windows targets that take the local-profile build
path and never consume the generated catalog. A transient models.dev
outage would therefore fail unrelated artifact builds.
Match the condition to the macOS signed release-profile build that
actually consumes the bundled catalog.
* fix(ci): embed built-in catalog in non-macOS native artifacts
The earlier narrowing to `runner.os == 'macOS' && inputs.sign-macos`
relied on a misread of build.mjs: its `profile === 'release'` guard
only auto-fetches the catalog as a dev fallback. Whether the binary
actually embeds the catalog is decided by tsdown's define at bundle
time, which reads KIMI_CODE_BUILT_IN_CATALOG_FILE regardless of
profile.
Linux and Windows release artifacts therefore lost their bundled
catalog and silently regressed offline /connect on those targets.
Restore generation for all OS jobs when sign-macos is true.
* chore(tui): mention /connect in welcome panel hints
Align the welcome panel with the /model picker so the empty-state copy
points users to both /login and /connect.
