@pentoshi/clai 3.9.7 → 3.11.0

package/dist/llm/provider.js

@@ -26,6 +26,17 @@ export const providerAliases = {
     "qwen-cloud": "qwen-cloud",
     dashscope: "qwen-cloud",
     qwencloud: "qwen-cloud",
+    modal: "modal",
+    "modal-labs": "modal",
+    modalcom: "modal",
+    lightning: "lightning",
+    "lightning-ai": "lightning",
+    lightningai: "lightning",
+    "lightning.ai": "lightning",
+    litai: "lightning",
+    tokenrouter: "tokenrouter",
+    "token-router": "tokenrouter",
+    tr: "tokenrouter",
 };
 export const defaultModels = {
     groq: "llama-3.3-70b-versatile",
@@ -40,6 +51,12 @@ export const defaultModels = {
     ollama: "llama3.1:8b",
     bynara: "mimo-v2.5-free",
     "qwen-cloud": "qwen3.7-plus",
+    // Modal Endpoints serve one model per endpoint, named by its source repo id.
+    modal: "moonshotai/Kimi-K3",
+    // Lightning AI namespaces model ids by vendor.
+    lightning: "openai/gpt-5",
+    // TokenRouter ids are short and case-sensitive (kimi-k2p6, not Kimi K2.6).
+    tokenrouter: "kimi-k2p6",
 };
 const retiredModelReplacements = {
     groq: {
@@ -78,6 +95,12 @@ export const envVars = {
     ollama: "OLLAMA_HOST",
     bynara: "BYNARA_API_KEY",
     "qwen-cloud": "DASHSCOPE_API_KEY",
+    // Modal needs a *pair* of proxy-token env vars plus an endpoint URL; the
+    // second half (MODAL_PROXY_TOKEN_SECRET) and MODAL_BASE_URL are resolved in
+    // store/keys.ts and store/config.ts respectively.
+    modal: "MODAL_PROXY_TOKEN_ID",
+    lightning: "LIGHTNING_API_KEY",
+    tokenrouter: "TOKENROUTER_API_KEY",
 };
 export function normalizeProvider(value) {
     return providerAliases[value.trim().toLowerCase()];
@@ -97,6 +120,28 @@ export function sanitizeProviderModel(provider, model) {
     const replacement = retiredModelReplacements[provider]?.[normalized.toLowerCase()];
     return replacement ?? normalized;
 }
+/**
+ * Normalize a user-supplied OpenAI-compatible base URL so the HTTP helpers can
+ * append `/chat/completions` to it. Idempotent, and tolerant of the three
+ * things people actually paste: a bare host, a trailing slash, or a full
+ * endpoint path.
+ *
+ *   x--ep-y.modal.direct                  → https://x--ep-y.modal.direct/v1
+ *   https://x--ep-y.modal.direct/v1/      → https://x--ep-y.modal.direct/v1
+ *   https://host/v1/chat/completions      → https://host/v1
+ */
+export function normalizeEndpointUrl(url) {
+    let value = url.trim();
+    if (!value)
+        return "";
+    if (!/^https?:\/\//i.test(value))
+        value = `https://${value}`;
+    value = value.replace(/\/+$/, "");
+    value = value.replace(/\/chat\/completions$/i, "").replace(/\/models$/i, "");
+    if (!/\/v\d+$/i.test(value))
+        value = `${value}/v1`;
+    return value;
+}
 export function maskSecret(secret) {
     // Show first 4 and last 4 characters with a fixed-width •••• separator.
     // Output is always 12 chars for keys >= 8, keeping tables compact.
@@ -120,9 +165,279 @@ export function redactSecrets(value) {
         .replace(/sk-[A-Za-z0-9._-]+/g, "sk-••••••")
         .replace(/sk-or-[A-Za-z0-9_-]+/g, "sk-or-••••••")
         .replace(/sk-ant-[A-Za-z0-9_-]+/g, "sk-ant-••••••")
-        .replace(/nvapi-[A-Za-z0-9_-]+/g, "nvapi-••••••");
+        .replace(/nvapi-[A-Za-z0-9_-]+/g, "nvapi-••••••")
+        // Modal proxy tokens: id `wk-…`, secret `ws-…`. Require a long tail so
+        // ordinary words (`ws-connection`) are never mangled.
+        .replace(/wk-[A-Za-z0-9_-]{16,}/g, "wk-••••••")
+        .replace(/ws-[A-Za-z0-9_-]{16,}/g, "ws-••••••");
 }
 export const providerInfo = {
+    tokenrouter: `TokenRouter — one key for frontier open models
+
+WHAT IT IS
+  An OpenAI-compatible gateway that fronts Kimi, DeepSeek, Qwen, GLM, GPT-OSS
+  and MiniMax behind a single bearer key and base URL. Chat Completions plus
+  the Responses API upstream; clai uses Chat Completions with SSE streaming,
+  native tool calling and JSON mode.
+
+  Base URL   https://api.tokenrouter.com/v1   (override if your account uses
+             a different host — see ENDPOINTS below)
+  Auth       Authorization: Bearer <key>
+  Endpoints  /models · /chat/completions
+
+MODELS (ids are exact and case-sensitive — kimi-k2p6, not "Kimi K2.6")
+  kimi-k2p7-code / -fast   256K ctx   agentic coding, tools + vision
+  kimi-k2p6                256K ctx   long-horizon agent work (clai default)
+  kimi-k2p5                256K ctx   faster, cheaper coding
+  deepseek-v4-pro          1M ctx     deep reasoning over huge codebases
+  deepseek-v4-flash        1M ctx     cheapest long context
+  qwen3p7-plus             256K ctx   multilingual general purpose
+  qwen3p6-plus             256K ctx   multilingual general purpose
+  glm-5p1 / glm-5p1-fast   200K ctx   agentic engineering
+  gpt-oss-120b             128K ctx   open-source OpenAI model
+  minimax-m3               512K ctx   very long context, agent harnesses
+  minimax-m2p7            196K ctx    budget agent tasks
+
+  /model reads the live list from /models, which is filtered to the channels
+  your key can actually reach — trust that over this table.
+
+COST
+  Prepaid balance billed per token; the dashboard header shows what is left.
+  There is no published free tier, so treat it as paid: clai classes it
+  paid-cloud and /freeonly on keeps it out of the fallback chain. Prices vary
+  a lot between the flash/fast variants and the pro ones — deepseek-v4-flash
+  is the cheap long-context option, kimi-k2p7-code the premium coding one.
+
+SETUP
+  1. Create an API key in your TokenRouter account under API Keys.
+  2. clai set tokenrouter sk-yourKey
+  3. clai use tokenrouter
+  4. /model kimi-k2p6      (or any id from /model)
+
+MANAGING KEYS AND ENDPOINTS IN CLAI
+  clai set tokenrouter <key>            add a key (up to 10, rotated on failure)
+  clai keys                             masked keys + the active endpoint
+  clai unset tokenrouter                remove every stored key
+  /set tokenrouter                      TUI: endpoint editor, then key editor
+  /info tokenrouter                     this page
+
+  Base URL (optional — defaults to api.tokenrouter.com):
+  clai set tokenrouter --url https://tokenrouter.me/v1
+  clai set tokenrouter --url <a> --url <b>   store several, sticky active one
+  clai unset tokenrouter --url               back to the default
+  TOKENROUTER_BASE_URL overrides the whole list.
+
+GOOD TO KNOW
+  - Reasoning models return their thinking in reasoning_content; clai folds it
+    into the usual thinking block, so /think and /variants behave normally.
+  - max_tokens above a model's ceiling is clamped by the gateway rather than
+    rejected, but prompt + max_tokens must still fit the context window.
+  - The model field in responses may echo a fully-qualified upstream path
+    instead of the short id you sent. That is expected.
+  - Env var: TOKENROUTER_API_KEY (used when nothing is stored).
+
+Docs: https://docs.tokenrouter.me`,
+    lightning: `Lightning AI Model APIs — one key for OpenAI, Anthropic, Google
+and Lightning-hosted open models
+
+WHAT IT IS
+  An OpenAI-compatible gateway at https://lightning.ai/api/v1 that fronts
+  frontier models from several vendors plus open models Lightning serves
+  itself. One account, one key, no per-vendor subscriptions, billed by the
+  token. Model ids are vendor-namespaced:
+    openai/gpt-5, openai/o3, openai/gpt-5.6-sol
+    anthropic/claude-opus-4-8, anthropic/claude-sonnet-4-6
+    google/gemini-3.5-flash, google/gemini-2.5-pro
+    lightning-ai/gpt-oss-120b, lightning-ai/deepseek-v4-pro,
+    lightning-ai/nemotron-3-ultra-550b-a55b, lightning-ai/gemma-4-31B-it
+
+FREE TIER — WHAT YOU GET
+  New accounts get up to 40 million free tokens to start, and the docs
+  advertise no subscription or credit card to begin. After that it is
+  pay-per-token at each model's rate — the /models endpoint returns the
+  per-token input and output price for every id, so nothing is hidden.
+
+  Rate limits by plan (requests/min · tokens/min):
+    Free         15 ·  120,000
+    Pro          20 ·  120,000
+    Teams        30 ·  150,000
+    Enterprise  300 ·  unlimited
+
+  Signing up needs a non-virtual phone number, and the free grant is once
+  per person — a second account does not get a second grant.
+
+SETUP — STEP BY STEP
+  1. Create an account at https://lightning.ai
+  2. Open the Model APIs page and reveal your key:
+       https://lightning.ai/lightning-ai/model-apis/models?showApiKey=true
+  3. Give it to clai:
+       clai set lightning <your-api-key>
+       clai use lightning
+  4. Pick a model — /model lists the live catalog from the gateway:
+       /model openai/gpt-5
+       /model lightning-ai/gpt-oss-120b     (cheapest open-weight option)
+
+  Or export it instead of storing it:
+       LIGHTNING_API_KEY=<your-api-key>
+
+MANAGING KEYS AND ENDPOINTS IN CLAI
+  clai set lightning <key>            add a key (up to 10, rotated on failure)
+  clai set lightning <key2>           add another; the last that worked is sticky
+  clai keys                           masked keys + the active endpoint
+  clai unset lightning                remove every stored key
+  /set lightning                      TUI: endpoint editor, then multi-key editor
+  /info lightning                     this page
+
+  Base URL (optional — defaults to the shared gateway above):
+  clai set lightning --url <url>       add an endpoint, make it active
+  clai set lightning --url <a> --url <b>   add several
+  clai unset lightning --url           back to the default gateway
+  /set lightning https://...           add + activate one endpoint
+
+  Point it at a private Lightning Inference deployment or a proxy that speaks
+  the same OpenAI routes; up to 10 URLs are stored with a sticky active one.
+  LIGHTNING_BASE_URL overrides the whole list.
+
+GOOD TO KNOW
+  - The catalog lists one entry per published agent/preset, so the same model
+    id appears more than once upstream; clai dedupes it for /model.
+  - Cost varies enormously across the list — claude-opus and gpt-5.x cost
+    dollars per million tokens while lightning-ai/gpt-oss-* are cents. Check
+    the price on https://lightning.ai/models before long agent runs.
+  - Streaming, native tool calling and reasoning_effort all work; clai sends
+    the standard OpenAI knobs and retries without them if a model objects.
+  - Classed as paid-cloud, so /freeonly on keeps it out of the fallback
+    chain even while the free tokens last.
+  - Model APIs are separate from Lightning Studios/GPU credits; the token
+    grant is not the same balance as Studio compute credits.
+
+Docs:   https://lightning.ai/docs/platform/inference/model-apis
+Models: https://lightning.ai/models
+API:    https://lightning.ai/api/v1 (OpenAI-compatible; /models, /chat/completions)`,
+    modal: `Modal Endpoints — your own serverless OpenAI-compatible endpoint
+
+WHAT IT IS
+  Modal Endpoints deploy an open-weight model (Kimi, Qwen, DeepSeek, GLM,
+  Gemma, GPT-OSS, Nemotron, or your own fine-tune) behind a low-latency
+  request proxy. The endpoint serves the standard Chat Completions API under
+  /v1, autoscales under load, and scales to zero when idle. The endpoint
+  belongs to your workspace, so the base URL is unique to you:
+    https://<workspace>--ep-<endpoint>.<region>.modal.direct/v1
+
+FREE TIER — WHAT YOU GET AND WHAT IT COSTS
+  Signing up is free. The Starter plan is $0/month + compute and includes
+  $30 of free compute credit every month; you unlock it by adding a payment
+  method, and nothing is charged until a month's usage passes the credit.
+  Starter also includes 3 workspace seats, 100 containers + 10 GPU
+  concurrency, region selection, and real-time metrics and logs.
+  Team is $250/month + compute with $100/month of credit, 1000 containers
+  and 50 GPU concurrency. Credit grants exist for early-stage startups and
+  for academics (up to $10k).
+
+  Billing is per-second compute (GPU + CPU) at standard Modal rates — not
+  per token. Because endpoints scale to zero you pay only while a container
+  is starting or serving; an idle endpoint costs nothing. Pinning compute to
+  a region applies a price multiplier. The credit is granted per month, so
+  keep an eye on the Credits figure in the dashboard header.
+
+SETUP — STEP BY STEP
+  1. Create a free account at https://modal.com
+  2. Add a payment method (dashboard → settings → billing) to activate the
+     $30/month credit.
+  3. Install the CLI and log in:
+       pip install modal
+       modal setup      (browser login; writes tokens to ~/.modal.toml)
+  4. Deploy an endpoint from the Endpoints tab in the dashboard, or:
+       modal endpoint create --model moonshotai/Kimi-K3 --name kimi-k3
+       modal endpoint create --model Qwen/Qwen3.5-4B --routing-region us-east
+     Provisioning takes a few minutes. Region default is us-west.
+  5. Copy the endpoint URL from the endpoint Overview page, or run
+       modal endpoint list
+  6. Create a proxy token pair (endpoints are authenticated by default):
+       modal workspace proxy-tokens create
+     This prints a token ID (wk-...) and a token secret (ws-...). THE SECRET
+     IS SHOWN ONLY ONCE — copy it now; it cannot be retrieved later. You can
+     also create one in the dashboard under workspace settings, and the
+     endpoint Quickstart panel lists the token IDs you already have.
+     On RBAC workspaces, also scope it to the endpoint's environment:
+       modal workspace proxy-tokens allow <token-id> <environment>
+  7. Point clai at it (URL from step 5, pair from step 6):
+       clai set modal --url <endpoint-url>
+       clai set modal wk-yourTokenId:ws-yourTokenSecret
+       clai use modal
+  8. Verify: "clai keys" shows the masked pair and the endpoint it points at,
+     and /model lists the models the endpoint actually serves.
+
+KEYS — WHAT IS NEEDED AND WHERE IT GOES
+  Endpoint URL        stored in config as modalBaseUrl (not a secret)
+  Proxy token ID      wk-...  sent as the Modal-Key header
+  Proxy token secret  ws-...  sent as the Modal-Secret header
+
+  There is no bearer API key. clai stores the pair as ONE secret shaped
+  "<token-id>:<token-secret>" in the OS keychain (or ~/.clai/keys.json with
+  restricted permissions when no keychain is available), so masking,
+  rotation and /unset behave exactly like every other provider.
+
+  Proxy tokens (wk-/ws-) are NOT Modal API tokens (ak-/as-). API tokens
+  authenticate the CLI and SDK; they will not work as endpoint headers.
+
+MANAGING ENDPOINTS AND KEYS IN CLAI
+  Both are multi-entry lists with a sticky active choice, up to 10 each.
+
+  clai set modal --url <endpoint>          add an endpoint, make it active
+  clai set modal --url <a> --url <b>       add several in one call
+  clai set modal --url <known endpoint>    re-activate one already stored
+  clai set modal wk-id:ws-secret           add a token pair
+  clai set modal                           prompt for the pair (input hidden)
+  clai keys                                every endpoint + masked pairs, ★ active
+  clai unset modal --url                   clear the endpoint list only
+  clai unset modal                         clear the token pairs only
+  /set modal                               TUI: endpoint editor, then keys
+  /set modal https://...                   add + activate one endpoint
+  /info modal                              this page
+
+  One endpoint serves one model, so keep an endpoint per model you deploy and
+  switch with clai set modal --url <that endpoint> — or star a row in the
+  /set endpoint editor. Endpoints are not auto-rotated on failure, because a
+  different endpoint serves a different model; token pairs DO rotate like any
+  other provider (401 / 403 / 429 / 5xx / empty response moves to the next).
+
+ENVIRONMENT VARIABLES (used only when nothing is stored)
+  MODAL_BASE_URL              endpoint URL; overrides the stored list entirely
+                              ("/v1" is appended if missing)
+  MODAL_PROXY_TOKEN_ID        wk-...  both halves needed, or it is ignored
+  MODAL_PROXY_TOKEN_SECRET    ws-...
+  MODAL_SESSION_ID            optional sticky-session id. When unset, clai
+                              generates one per run so a whole conversation
+                              lands on the same warm container.
+
+GOOD TO KNOW
+  - Cold start: the first request after idle pays container start-up, which
+    can take tens of seconds on a large model. clai allows up to 3 minutes
+    for the first token before treating a stream as stalled.
+  - The model name on the wire is the source repo id (e.g.
+    moonshotai/Kimi-K3), not a Modal alias. /model reads the live list.
+  - Endpoint URLs are stored as a list, so several deployments can live side
+    by side; only the active one is used. "clai keys" shows them all.
+  - Streaming, native tool calling, structured outputs and thinking all work.
+    /variants on|off maps to Modal's reasoning toggle.
+  - Finished experimenting? "modal endpoint stop <name>" tears the endpoint
+    down so no stray request can wake it and spend credit.
+  - Modal counts as paid-cloud, so /freeonly on keeps it out of the
+    cross-provider fallback chain.
+
+TROUBLESHOOTING
+  401 / 403         wrong, revoked or unscoped pair — or an API token (ak-)
+                    was pasted instead of a proxy token (wk-)
+  404               endpoint URL is wrong, or the endpoint was stopped
+  "endpoint URL is not configured"
+                    run clai set modal --url <endpoint>
+  model not found   send the repo id reported by /model or modal endpoint list
+  slow first token  cold start; the endpoint had scaled to zero
+
+Docs:    https://modal.com/docs/guide/endpoints
+Auth:    https://modal.com/docs/guide/webhook-proxy-auth
+Pricing: https://modal.com/pricing`,
     bynara: `Current Plan
 
 Free

package/dist/llm/provider.js.map

@@ -1 +1 @@
-{"version":3,"file":"provider.js","sourceRoot":"","sources":["../../src/llm/provider.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AA0B1C,MAAM,CAAC,MAAM,eAAe,GAA+B;IACzD,IAAI,EAAE,MAAM;IACZ,MAAM,EAAE,QAAQ;IAChB,MAAM,EAAE,QAAQ;IAChB,UAAU,EAAE,YAAY;IACxB,MAAM,EAAE,QAAQ;IAChB,SAAS,EAAE,WAAW;IACtB,MAAM,EAAE,WAAW;IACnB,MAAM,EAAE,QAAQ;IAChB,GAAG,EAAE,QAAQ;IACb,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,aAAa;IAC1B,cAAc,EAAE,aAAa;IAC7B,MAAM,EAAE,aAAa;IACrB,MAAM,EAAE,QAAQ;IAChB,MAAM,EAAE,QAAQ;IAChB,YAAY,EAAE,YAAY;IAC1B,MAAM,EAAE,QAAQ;IAChB,KAAK,EAAE,QAAQ;IACf,MAAM,EAAE,QAAQ;IAChB,eAAe,EAAE,QAAQ;IACzB,UAAU,EAAE,QAAQ;IACpB,IAAI,EAAE,QAAQ;IACd,IAAI,EAAE,YAAY;IAClB,YAAY,EAAE,YAAY;IAC1B,SAAS,EAAE,YAAY;IACvB,SAAS,EAAE,YAAY;CACxB,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAA+B;IACvD,IAAI,EAAE,yBAAyB;IAC/B,MAAM,EAAE,kBAAkB;IAC1B,UAAU,EAAE,wCAAwC;IACpD,MAAM,EAAE,cAAc;IACtB,SAAS,EAAE,yBAAyB;IACpC,MAAM,EAAE,oBAAoB;IAC5B,WAAW,EAAE,iBAAiB;IAC9B,MAAM,EAAE,WAAW;IACnB,YAAY,EAAE,4BAA4B;IAC1C,MAAM,EAAE,aAAa;IACrB,MAAM,EAAE,gBAAgB;IACxB,YAAY,EAAE,cAAc;CAC7B,CAAC;AAEF,MAAM,wBAAwB,GAAwD;IACpF,IAAI,EAAE;QACJ,cAAc,EAAE,sBAAsB;QACtC,6BAA6B,EAAE,qBAAqB;QACpD,+BAA+B,EAAE,yBAAyB;QAC1D,iBAAiB,EAAE,yBAAyB;QAC5C,gBAAgB,EAAE,sBAAsB;QACxC,+CAA+C,EAC7C,2CAA2C;KAC9C;IACD,MAAM,EAAE;QACN,kBAAkB,EAAE,kBAAkB;QACtC,uBAAuB,EAAE,uBAAuB;KACjD;IACD,MAAM,EAAE;QACN,yEAAyE;QACzE,0DAA0D;QAC1D,wCAAwC,EAAE,aAAa,CAAC,MAAM;KAC/D;IACD,MAAM,EAAE;QACN,4DAA4D;QAC5D,aAAa,EAAE,cAAc;QAC7B,QAAQ,EAAE,SAAS;KACpB;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,OAAO,GAA2C;IAC7D,IAAI,EAAE,cAAc;IACpB,MAAM,EAAE,gBAAgB;IACxB,UAAU,EAAE,oBAAoB;IAChC,MAAM,EAAE,gBAAgB;IACxB,SAAS,EAAE,mBAAmB;IAC9B,MAAM,EAAE,gBAAgB;IACxB,WAAW,EAAE,qBAAqB;IAClC,MAAM,EAAE,gBAAgB;IACxB,YAAY,EAAE,mBAAmB;IACjC,MAAM,EAAE,aAAa;IACrB,MAAM,EAAE,gBAAgB;IACxB,YAAY,EAAE,mBAAmB;CAClC,CAAC;AAEF,MAAM,UAAU,iBAAiB,CAAC,KAAa;IAC7C,OAAO,eAAe,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AACrD,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,KAAa;IAC1C,MAAM,QAAQ,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;IAC1C,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CACb,yBAAyB,KAAK,2BAA2B,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAClF,CAAC;IACJ,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,QAAoB;IAClD,OAAO,aAAa,CAAC,QAAQ,CAAC,CAAC;AACjC,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,QAAoB,EAAE,KAAa;IACvE,MAAM,UAAU,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;IAChC,MAAM,WAAW,GACf,wBAAwB,CAAC,QAAQ,CAAC,EAAE,CAAC,UAAU,CAAC,WAAW,EAAE,CAAC,CAAC;IACjE,OAAO,WAAW,IAAI,UAAU,CAAC;AACnC,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,MAAc;IACvC,wEAAwE;IACxE,mEAAmE;IACnE,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,UAAU,CAAC;IAC7B,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AACxD,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,cAAc,CAAC,MAAc;IAC3C,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,MAAM,CAAC;IACzB,OAAO,IAAI,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,KAAa;IACzC,OAAO,KAAK;SACT,OAAO,CAAC,qBAAqB,EAAE,YAAY,CAAC;SAC5C,OAAO,CAAC,qBAAqB,EAAE,YAAY,CAAC;SAC5C,OAAO,CAAC,qBAAqB,EAAE,WAAW,CAAC;SAC3C,OAAO,CAAC,qBAAqB,EAAE,WAAW,CAAC;SAC3C,OAAO,CAAC,uBAAuB,EAAE,cAAc,CAAC;SAChD,OAAO,CAAC,wBAAwB,EAAE,eAAe,CAAC;SAClD,OAAO,CAAC,uBAAuB,EAAE,cAAc,CAAC,CAAC;AACtD,CAAC;AAED,MAAM,CAAC,MAAM,YAAY,GAA2B;IAClD,MAAM,EAAE;;;;;;;;;;;;;;;UAeA;CACT,CAAC;AAEF,MAAM,UAAU,mBAAmB,CAAC,QAAgB;IAClD,OAAO,YAAY,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,IAAI,mBAAmB,CAAC;AACrE,CAAC"}
+{"version":3,"file":"provider.js","sourceRoot":"","sources":["../../src/llm/provider.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AA0B1C,MAAM,CAAC,MAAM,eAAe,GAA+B;IACzD,IAAI,EAAE,MAAM;IACZ,MAAM,EAAE,QAAQ;IAChB,MAAM,EAAE,QAAQ;IAChB,UAAU,EAAE,YAAY;IACxB,MAAM,EAAE,QAAQ;IAChB,SAAS,EAAE,WAAW;IACtB,MAAM,EAAE,WAAW;IACnB,MAAM,EAAE,QAAQ;IAChB,GAAG,EAAE,QAAQ;IACb,IAAI,EAAE,QAAQ;IACd,WAAW,EAAE,aAAa;IAC1B,cAAc,EAAE,aAAa;IAC7B,MAAM,EAAE,aAAa;IACrB,MAAM,EAAE,QAAQ;IAChB,MAAM,EAAE,QAAQ;IAChB,YAAY,EAAE,YAAY;IAC1B,MAAM,EAAE,QAAQ;IAChB,KAAK,EAAE,QAAQ;IACf,MAAM,EAAE,QAAQ;IAChB,eAAe,EAAE,QAAQ;IACzB,UAAU,EAAE,QAAQ;IACpB,IAAI,EAAE,QAAQ;IACd,IAAI,EAAE,YAAY;IAClB,YAAY,EAAE,YAAY;IAC1B,SAAS,EAAE,YAAY;IACvB,SAAS,EAAE,YAAY;IACvB,KAAK,EAAE,OAAO;IACd,YAAY,EAAE,OAAO;IACrB,QAAQ,EAAE,OAAO;IACjB,SAAS,EAAE,WAAW;IACtB,cAAc,EAAE,WAAW;IAC3B,WAAW,EAAE,WAAW;IACxB,cAAc,EAAE,WAAW;IAC3B,KAAK,EAAE,WAAW;IAClB,WAAW,EAAE,aAAa;IAC1B,cAAc,EAAE,aAAa;IAC7B,EAAE,EAAE,aAAa;CAClB,CAAC;AAEF,MAAM,CAAC,MAAM,aAAa,GAA+B;IACvD,IAAI,EAAE,yBAAyB;IAC/B,MAAM,EAAE,kBAAkB;IAC1B,UAAU,EAAE,wCAAwC;IACpD,MAAM,EAAE,cAAc;IACtB,SAAS,EAAE,yBAAyB;IACpC,MAAM,EAAE,oBAAoB;IAC5B,WAAW,EAAE,iBAAiB;IAC9B,MAAM,EAAE,WAAW;IACnB,YAAY,EAAE,4BAA4B;IAC1C,MAAM,EAAE,aAAa;IACrB,MAAM,EAAE,gBAAgB;IACxB,YAAY,EAAE,cAAc;IAC5B,6EAA6E;IAC7E,KAAK,EAAE,oBAAoB;IAC3B,+CAA+C;IAC/C,SAAS,EAAE,cAAc;IACzB,2EAA2E;IAC3E,WAAW,EAAE,WAAW;CACzB,CAAC;AAEF,MAAM,wBAAwB,GAAwD;IACpF,IAAI,EAAE;QACJ,cAAc,EAAE,sBAAsB;QACtC,6BAA6B,EAAE,qBAAqB;QACpD,+BAA+B,EAAE,yBAAyB;QAC1D,iBAAiB,EAAE,yBAAyB;QAC5C,gBAAgB,EAAE,sBAAsB;QACxC,+CAA+C,EAC7C,2CAA2C;KAC9C;IACD,MAAM,EAAE;QACN,kBAAkB,EAAE,kBAAkB;QACtC,uBAAuB,EAAE,uBAAuB;KACjD;IACD,MAAM,EAAE;QACN,yEAAyE;QACzE,0DAA0D;QAC1D,wCAAwC,EAAE,aAAa,CAAC,MAAM;KAC/D;IACD,MAAM,EAAE;QACN,4DAA4D;QAC5D,aAAa,EAAE,cAAc;QAC7B,QAAQ,EAAE,SAAS;KACpB;CACF,CAAC;AAEF,MAAM,CAAC,MAAM,OAAO,GAA2C;IAC7D,IAAI,EAAE,cAAc;IACpB,MAAM,EAAE,gBAAgB;IACxB,UAAU,EAAE,oBAAoB;IAChC,MAAM,EAAE,gBAAgB;IACxB,SAAS,EAAE,mBAAmB;IAC9B,MAAM,EAAE,gBAAgB;IACxB,WAAW,EAAE,qBAAqB;IAClC,MAAM,EAAE,gBAAgB;IACxB,YAAY,EAAE,mBAAmB;IACjC,MAAM,EAAE,aAAa;IACrB,MAAM,EAAE,gBAAgB;IACxB,YAAY,EAAE,mBAAmB;IACjC,yEAAyE;IACzE,4EAA4E;IAC5E,kDAAkD;IAClD,KAAK,EAAE,sBAAsB;IAC7B,SAAS,EAAE,mBAAmB;IAC9B,WAAW,EAAE,qBAAqB;CACnC,CAAC;AAEF,MAAM,UAAU,iBAAiB,CAAC,KAAa;IAC7C,OAAO,eAAe,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;AACrD,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,KAAa;IAC1C,MAAM,QAAQ,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;IAC1C,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CACb,yBAAyB,KAAK,2BAA2B,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAClF,CAAC;IACJ,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,QAAoB;IAClD,OAAO,aAAa,CAAC,QAAQ,CAAC,CAAC;AACjC,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,QAAoB,EAAE,KAAa;IACvE,MAAM,UAAU,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;IAChC,MAAM,WAAW,GACf,wBAAwB,CAAC,QAAQ,CAAC,EAAE,CAAC,UAAU,CAAC,WAAW,EAAE,CAAC,CAAC;IACjE,OAAO,WAAW,IAAI,UAAU,CAAC;AACnC,CAAC;AAED;;;;;;;;;GASG;AACH,MAAM,UAAU,oBAAoB,CAAC,GAAW;IAC9C,IAAI,KAAK,GAAG,GAAG,CAAC,IAAI,EAAE,CAAC;IACvB,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IACtB,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,KAAK,GAAG,WAAW,KAAK,EAAE,CAAC;IAC7D,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAClC,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,uBAAuB,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC;IAC7E,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC;QAAE,KAAK,GAAG,GAAG,KAAK,KAAK,CAAC;IACnD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,MAAc;IACvC,wEAAwE;IACxE,mEAAmE;IACnE,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,UAAU,CAAC;IAC7B,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AACxD,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,cAAc,CAAC,MAAc;IAC3C,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,MAAM,CAAC;IACzB,OAAO,IAAI,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,KAAa;IACzC,OAAO,KAAK;SACT,OAAO,CAAC,qBAAqB,EAAE,YAAY,CAAC;SAC5C,OAAO,CAAC,qBAAqB,EAAE,YAAY,CAAC;SAC5C,OAAO,CAAC,qBAAqB,EAAE,WAAW,CAAC;SAC3C,OAAO,CAAC,qBAAqB,EAAE,WAAW,CAAC;SAC3C,OAAO,CAAC,uBAAuB,EAAE,cAAc,CAAC;SAChD,OAAO,CAAC,wBAAwB,EAAE,eAAe,CAAC;SAClD,OAAO,CAAC,uBAAuB,EAAE,cAAc,CAAC;QACjD,uEAAuE;QACvE,sDAAsD;SACrD,OAAO,CAAC,wBAAwB,EAAE,WAAW,CAAC;SAC9C,OAAO,CAAC,wBAAwB,EAAE,WAAW,CAAC,CAAC;AACpD,CAAC;AAED,MAAM,CAAC,MAAM,YAAY,GAA2B;IAClD,WAAW,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kCAgEmB;IAChC,SAAS,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;oFA4EuE;IAClF,KAAK,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;mCA2H0B;IACjC,MAAM,EAAE;;;;;;;;;;;;;;;UAeA;CACT,CAAC;AAEF,MAAM,UAAU,mBAAmB,CAAC,QAAgB;IAClD,OAAO,YAAY,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,IAAI,mBAAmB,CAAC;AACrE,CAAC"}

package/dist/llm/qwen-cloud.js

@@ -1,5 +1,5 @@
 import { defaultModels, } from "./provider.js";
-import { openAiCompatibleComplete, openAiCompatiblePing, openAiCompatibleStream, toCompletionResult, readJson, } from "./http.js";
+import { openAiCompatibleComplete, openAiCompatiblePing, openAiCompatibleStream, toCompletionResult, readJson, ingestOpenAiModelCatalog, } from "./http.js";
 // Qwen Cloud documents this OpenAI-compatible endpoint for international
 // accounts. See https://docs.qwencloud.com/developer-guides/getting-started/first-api-call
 const baseUrl = "https://dashscope-intl.aliyuncs.com/compatible-mode/v1";
@@ -25,10 +25,7 @@ export const qwenCloudProvider = {
             headers: { authorization: `Bearer ${auth.apiKey}` },
         });
         const data = await readJson(response);
-        const models = (data.data ?? [])
-            .map((model) => model.id)
-            .filter((model) => Boolean(model))
-            .sort();
+        const models = ingestOpenAiModelCatalog("qwen-cloud", data);
         if (models.length > 0) {
             cachedModels = models;
             lastFetchTime = now;

package/dist/llm/qwen-cloud.js.map

@@ -1 +1 @@
-{"version":3,"file":"qwen-cloud.js","sourceRoot":"","sources":["../../src/llm/qwen-cloud.ts"],"names":[],"mappings":"AACA,OAAO,EACL,aAAa,GAGd,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,wBAAwB,EACxB,oBAAoB,EACpB,sBAAsB,EACtB,kBAAkB,EAClB,QAAQ,GACT,MAAM,WAAW,CAAC;AAEnB,yEAAyE;AACzE,2FAA2F;AAC3F,MAAM,OAAO,GAAG,wDAAwD,CAAC;AAEzE,IAAI,YAAY,GAAoB,IAAI,CAAC;AACzC,IAAI,aAAa,GAAG,CAAC,CAAC;AACtB,MAAM,YAAY,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AAEpC,MAAM,CAAC,MAAM,iBAAiB,GAAgB;IAC5C,EAAE,EAAE,YAAY;IAChB,WAAW,EAAE,YAAY;IACzB,YAAY,EAAE,aAAa,CAAC,YAAY,CAAC;IACzC,MAAM,EAAE,mBAAmB;IAC3B,uEAAuE;IACvE,iDAAiD;IACjD,WAAW,EAAE,CAAC,GAAW,EAAE,EAAE,CAAC,yBAAyB,CAAC,IAAI,CAAC,GAAG,CAAC;IACjE,KAAK,CAAC,UAAU,CAAC,IAAkB;QACjC,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,IAAI,YAAY,IAAI,GAAG,GAAG,aAAa,GAAG,YAAY,EAAE,CAAC;YACvD,OAAO,YAAY,CAAC;QACtB,CAAC;QACD,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,SAAS,EAAE;YAChD,OAAO,EAAE,EAAE,aAAa,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE,EAAE;SACpD,CAAC,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAoC,QAAQ,CAAC,CAAC;QACzE,MAAM,MAAM,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC;aAC7B,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC;aACxB,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;aAClD,IAAI,EAAE,CAAC;QACV,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,YAAY,GAAG,MAAM,CAAC;YACtB,aAAa,GAAG,GAAG,CAAC;QACtB,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,IAAkB;QAC3B,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,oBAAoB,CAAC,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;IACnD,CAAC;IACD,KAAK,CAAC,QAAQ,CACZ,OAA0B,EAC1B,IAAkB;QAElB,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,aAAa,CAAC,YAAY,CAAC,CAAC;QAC3D,MAAM,OAAO,GAAG,MAAM,wBAAwB,CAAC;YAC7C,QAAQ,EAAE,YAAY;YACtB,UAAU,EAAE,YAAY;YACxB,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,KAAK;YACL,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,SAAS,EAAE,OAAO,CAAC,QAAQ;YAC3B,cAAc,EAAE,QAAQ;YACxB,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;SAC7C,CAAC,CAAC;QACH,OAAO,kBAAkB,CAAC,YAAY,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;IAC1D,CAAC;IACD,KAAK,CAAC,MAAM,CACV,OAA0B,EAC1B,IAAkB,EAClB,OAAgC;QAEhC,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,aAAa,CAAC,YAAY,CAAC,CAAC;QAC3D,MAAM,OAAO,GAAG,MAAM,sBAAsB,CAAC;YAC3C,QAAQ,EAAE,YAAY;YACtB,UAAU,EAAE,YAAY;YACxB,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,KAAK;YACL,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,OAAO;YACP,eAAe,EAAE,OAAO,CAAC,eAAe;YACxC,SAAS,EAAE,OAAO,CAAC,QAAQ;YAC3B,cAAc,EAAE,QAAQ;YACxB,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;SAC7C,CAAC,CAAC;QACH,OAAO,kBAAkB,CAAC,YAAY,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;IAC1D,CAAC;CACF,CAAC"}
+{"version":3,"file":"qwen-cloud.js","sourceRoot":"","sources":["../../src/llm/qwen-cloud.ts"],"names":[],"mappings":"AACA,OAAO,EACL,aAAa,GAGd,MAAM,eAAe,CAAC;AACvB,OAAO,EACL,wBAAwB,EACxB,oBAAoB,EACpB,sBAAsB,EACtB,kBAAkB,EAClB,QAAQ,EACR,wBAAwB,GACzB,MAAM,WAAW,CAAC;AAEnB,yEAAyE;AACzE,2FAA2F;AAC3F,MAAM,OAAO,GAAG,wDAAwD,CAAC;AAEzE,IAAI,YAAY,GAAoB,IAAI,CAAC;AACzC,IAAI,aAAa,GAAG,CAAC,CAAC;AACtB,MAAM,YAAY,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AAEpC,MAAM,CAAC,MAAM,iBAAiB,GAAgB;IAC5C,EAAE,EAAE,YAAY;IAChB,WAAW,EAAE,YAAY;IACzB,YAAY,EAAE,aAAa,CAAC,YAAY,CAAC;IACzC,MAAM,EAAE,mBAAmB;IAC3B,uEAAuE;IACvE,iDAAiD;IACjD,WAAW,EAAE,CAAC,GAAW,EAAE,EAAE,CAAC,yBAAyB,CAAC,IAAI,CAAC,GAAG,CAAC;IACjE,KAAK,CAAC,UAAU,CAAC,IAAkB;QACjC,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,IAAI,YAAY,IAAI,GAAG,GAAG,aAAa,GAAG,YAAY,EAAE,CAAC;YACvD,OAAO,YAAY,CAAC;QACtB,CAAC;QACD,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,SAAS,EAAE;YAChD,OAAO,EAAE,EAAE,aAAa,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE,EAAE;SACpD,CAAC,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAoC,QAAQ,CAAC,CAAC;QACzE,MAAM,MAAM,GAAG,wBAAwB,CAAC,YAAY,EAAE,IAAI,CAAC,CAAC;QAC5D,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtB,YAAY,GAAG,MAAM,CAAC;YACtB,aAAa,GAAG,GAAG,CAAC;QACtB,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,IAAkB;QAC3B,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,oBAAoB,CAAC,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC;IACnD,CAAC;IACD,KAAK,CAAC,QAAQ,CACZ,OAA0B,EAC1B,IAAkB;QAElB,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,aAAa,CAAC,YAAY,CAAC,CAAC;QAC3D,MAAM,OAAO,GAAG,MAAM,wBAAwB,CAAC;YAC7C,QAAQ,EAAE,YAAY;YACtB,UAAU,EAAE,YAAY;YACxB,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,KAAK;YACL,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,SAAS,EAAE,OAAO,CAAC,QAAQ;YAC3B,cAAc,EAAE,QAAQ;YACxB,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;SAC7C,CAAC,CAAC;QACH,OAAO,kBAAkB,CAAC,YAAY,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;IAC1D,CAAC;IACD,KAAK,CAAC,MAAM,CACV,OAA0B,EAC1B,IAAkB,EAClB,OAAgC;QAEhC,IAAI,CAAC,IAAI,CAAC,MAAM;YAAE,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;QACpE,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,aAAa,CAAC,YAAY,CAAC,CAAC;QAC3D,MAAM,OAAO,GAAG,MAAM,sBAAsB,CAAC;YAC3C,QAAQ,EAAE,YAAY;YACtB,UAAU,EAAE,YAAY;YACxB,OAAO;YACP,MAAM,EAAE,IAAI,CAAC,MAAM;YACnB,KAAK;YACL,QAAQ,EAAE,OAAO,CAAC,QAAQ;YAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;YAC5B,WAAW,EAAE,OAAO,CAAC,WAAW;YAChC,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,OAAO;YACP,eAAe,EAAE,OAAO,CAAC,eAAe;YACxC,SAAS,EAAE,OAAO,CAAC,QAAQ;YAC3B,cAAc,EAAE,QAAQ;YACxB,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,iBAAiB,EAAE,OAAO,CAAC,iBAAiB;SAC7C,CAAC,CAAC;QACH,OAAO,kBAAkB,CAAC,YAAY,EAAE,KAAK,EAAE,OAAO,CAAC,CAAC;IAC1D,CAAC;CACF,CAAC"}

package/dist/llm/router.js

@@ -1,10 +1,10 @@
-import { getConfig, providerCategory } from "../store/config.js";
+import { getActiveProviderEndpoint, getConfig, providerCategory, providerUsesEndpoints, } from "../store/config.js";
 import { getProviderKeys, getProviderSecret, markProviderKeySuccess, } from "../store/keys.js";
 import { anthropicProvider } from "./anthropic.js";
 import { geminiProvider } from "./gemini.js";
 import { groqProvider } from "./groq.js";
-import { ProviderError, isReasoningUnsupportedError } from "./http.js";
-import { markReasoningUnsupported } from "./capabilities.js";
+import { ProviderError, isImageInputUnsupportedError, isReasoningUnsupportedError, stripImagesFromMessages, } from "./http.js";
+import { learnModelVisionCapability, markModelUnavailable, markReasoningUnsupported, modelAcceptsImages, visionSubstitutionOrigin, } from "./capabilities.js";
 import { markStreamEmittedBytes, streamAlreadyEmitted, streamEmittedBytes, } from "./stream-progress.js";
 import { attemptsPerKey, buildKeyAttemptPlan, formatKeyEventStatus, isImmediateKeySwitchError, isKeyCircleStopError, isKeyRotatableError, isQuotaKeyError, } from "./key-rotation.js";
 import { nvidiaProvider } from "./nvidia.js";
@@ -16,6 +16,9 @@ import { ollamaProvider } from "./ollama.js";
 import { openaiProvider } from "./openai.js";
 import { openrouterProvider } from "./openrouter.js";
 import { qwenCloudProvider } from "./qwen-cloud.js";
+import { modalProvider } from "./modal.js";
+import { lightningProvider } from "./lightning.js";
+import { tokenrouterProvider } from "./tokenrouter.js";
 import { maskSecretTail } from "./provider.js";
 import { isToolsUnsupportedError, markTextOnlyModel, } from "./tool-protocol.js";
 const MAX_RETRIES = 6;
@@ -243,6 +246,9 @@ export const providers = {
     ollama: ollamaProvider,
     bynara: bynaraProvider,
     "qwen-cloud": qwenCloudProvider,
+    modal: modalProvider,
+    lightning: lightningProvider,
+    tokenrouter: tokenrouterProvider,
 };
 const fallbackOrder = [
     "nvidia",
@@ -257,6 +263,9 @@ const fallbackOrder = [
     "aws-mantle",
     "ollama",
     "qwen-cloud",
+    "modal",
+    "lightning",
+    "tokenrouter",
 ];
 export function buildFallbackChain(requested, freeOnly, enabled = false) {
     if (!enabled)
@@ -274,12 +283,22 @@ export async function providerAuth(provider) {
     if (provider === "ollama") {
         return { baseUrl: secret.value };
     }
+    // Endpoint providers carry both: the stored secret plus the active endpoint
+    // URL from config (Modal requires one; Lightning treats it as an override).
+    if (providerUsesEndpoints(provider)) {
+        const baseUrl = getActiveProviderEndpoint(provider);
+        return { apiKey: secret.value, ...(baseUrl ? { baseUrl } : {}) };
+    }
     return { apiKey: secret.value };
 }
 function authForSlot(providerId, value) {
     if (providerId === "ollama") {
         return { baseUrl: value };
     }
+    if (providerUsesEndpoints(providerId)) {
+        const baseUrl = getActiveProviderEndpoint(providerId);
+        return { apiKey: value, ...(baseUrl ? { baseUrl } : {}) };
+    }
     return { apiKey: value };
 }
 function makeKeyEmitter(onStatus, onKeyEvent) {
@@ -298,7 +317,11 @@ function makeKeyEmitter(onStatus, onKeyEvent) {
 async function tryCompleteOnce(provider, providerId, request, model, auth) {
     const activeRequest = { ...request, provider: providerId, model };
     try {
-        return await provider.complete(activeRequest, auth);
+        const result = await provider.complete(activeRequest, auth);
+        if (hasImageInput(activeRequest)) {
+            learnModelVisionCapability(providerId, model, true);
+        }
+        return result;
     }
     catch (error) {
         if (activeRequest.tools?.length && isToolsUnsupportedError(error)) {
@@ -317,9 +340,57 @@ async function tryCompleteOnce(provider, providerId, request, model, auth) {
             markReasoningUnsupported(model);
             return await provider.complete(activeRequest, auth);
         }
+        if (hasImageInput(activeRequest) && isImageInputUnsupportedError(error)) {
+            learnModelVisionCapability(providerId, model, false);
+            return await provider.complete(withoutImages(activeRequest), auth);
+        }
+        const restored = revertVisionSubstitution(providerId, model, activeRequest, error);
+        if (restored) {
+            return await provider.complete(restored.request, auth);
+        }
         throw error;
     }
 }
+function hasImageInput(request) {
+    return request.messages.some((message) => message.images?.length);
+}
+function withoutImages(request) {
+    return { ...request, messages: stripImagesFromMessages(request.messages) };
+}
+function isModelNotFoundError(error) {
+    const status = error && typeof error === "object" && "status" in error
+        ? Number(error.status)
+        : undefined;
+    if (status !== 404 && status !== 400)
+        return false;
+    const body = error && typeof error === "object" && "body" in error
+        ? String(error.body ?? "")
+        : "";
+    const message = error instanceof Error ? error.message : String(error);
+    const hay = `${message}\n${body}`.toLowerCase();
+    if (status === 404)
+        return true;
+    return /model[_ ]?not[_ ]?found|no such model|unknown model|model does not exist|invalid model|unavailable[- ]model/.test(hay);
+}
+function revertVisionSubstitution(providerId, model, request, error) {
+    if (!isModelNotFoundError(error))
+        return undefined;
+    const original = visionSubstitutionOrigin(providerId, model);
+    if (!original)
+        return undefined;
+    markModelUnavailable(providerId, model);
+    const keepImages = modelAcceptsImages(providerId, original);
+    return {
+        original,
+        request: {
+            ...request,
+            model: original,
+            messages: keepImages
+                ? request.messages
+                : stripImagesFromMessages(request.messages),
+        },
+    };
+}
 async function tryStreamOnce(provider, providerId, request, model, auth, onToken, onStatus) {
     const activeRequest = { ...request, provider: providerId, model };
     // Count what actually reached the caller so no layer above can
@@ -329,12 +400,20 @@ async function tryStreamOnce(provider, providerId, request, model, auth, onToken
         emittedBytes += token.length;
         onToken(token);
     };
+    const learnVisionOnSuccess = () => {
+        if (hasImageInput(activeRequest)) {
+            learnModelVisionCapability(providerId, model, true);
+        }
+    };
     try {
         if (provider.stream) {
-            return await provider.stream(activeRequest, auth, emit);
+            const streamed = await provider.stream(activeRequest, auth, emit);
+            learnVisionOnSuccess();
+            return streamed;
         }
         const result = await provider.complete(activeRequest, auth);
         emit(result.text);
+        learnVisionOnSuccess();
         return result;
     }
     catch (error) {
@@ -380,6 +459,41 @@ async function tryStreamOnce(provider, providerId, request, model, auth, onToken
                 throw markStreamEmittedBytes(retryError, emittedBytes);
             }
         }
+        if (emittedBytes === 0 &&
+            hasImageInput(activeRequest) &&
+            isImageInputUnsupportedError(error)) {
+            learnModelVisionCapability(providerId, model, false);
+            onStatus?.(`ℹ ${providerId}/${model} rejected image input — retrying without the attached image(s)`);
+            const textOnlyRequest = withoutImages(activeRequest);
+            try {
+                if (provider.stream) {
+                    return await provider.stream(textOnlyRequest, auth, emit);
+                }
+                const result = await provider.complete(textOnlyRequest, auth);
+                emit(result.text);
+                return result;
+            }
+            catch (retryError) {
+                throw markStreamEmittedBytes(retryError, emittedBytes);
+            }
+        }
+        if (emittedBytes === 0) {
+            const restored = revertVisionSubstitution(providerId, model, activeRequest, error);
+            if (restored) {
+                onStatus?.(`ℹ ${providerId}/${model} is not available on this account — falling back to ${restored.original}`);
+                try {
+                    if (provider.stream) {
+                        return await provider.stream(restored.request, auth, emit);
+                    }
+                    const result = await provider.complete(restored.request, auth);
+                    emit(result.text);
+                    return result;
+                }
+                catch (retryError) {
+                    throw markStreamEmittedBytes(retryError, emittedBytes);
+                }
+            }
+        }
         throw markStreamEmittedBytes(error, emittedBytes);
     }
 }
@@ -601,9 +715,15 @@ export async function streamWithProvider(request, onToken, onStatusOrOptions) {
 }
 export async function pingProvider(providerId, secretOverride) {
     const provider = providers[providerId];
+    const resolved = await providerAuth(providerId);
     const auth = providerId === "ollama"
-        ? { baseUrl: secretOverride ?? (await providerAuth(providerId)).baseUrl }
-        : { apiKey: secretOverride ?? (await providerAuth(providerId)).apiKey };
+        ? { baseUrl: secretOverride ?? resolved.baseUrl }
+        : providerUsesEndpoints(providerId)
+            ? {
+                apiKey: secretOverride ?? resolved.apiKey,
+                ...(resolved.baseUrl ? { baseUrl: resolved.baseUrl } : {}),
+            }
+            : { apiKey: secretOverride ?? resolved.apiKey };
     await provider.ping(auth);
 }
 //# sourceMappingURL=router.js.map