Improve MiniMax/Kimi provider ergonomics + stabilize slides/progress output

PR_DRAFT.md

@@ -0,0 +1,91 @@
+# Title
+Add first-class Kimi model support + stabilize slides/progress output
+
+## Summary
+This PR improves model/provider ergonomics and fixes multiple output stability issues seen in slides mode.
+
+### 1) First-class Kimi support (similar to MiniMax)
+- Add `--model kimi` alias (defaults to `openai/kimi-k2.5`)
+- Add `--model kimi/<model>`
+- Add provider env support:
+  - `KIMI_API_KEY` (primary)
+  - `MOONSHOT_API_KEY` (alias)
+  - `KIMI_BASE_URL` (primary)
+  - `MOONSHOT_BASE_URL` (alias)
+  - default base URL: `https://api.moonshot.ai/v1`
+- Wire Kimi handling through:
+  - model parsing/spec
+  - auto model candidate resolution
+  - env/runtime context
+  - URL and asset flows
+  - daemon agent/flow context
+  - help/docs
+
+### 2) Kimi runtime compatibility fix
+- Kimi rejects custom temperatures in this integration (`only 1 is allowed`).
+- For `openai/kimi...` models, force effective `temperature=1` to avoid runtime failures.
+
+### 3) Slides/progress output stabilization
+- Fix progress gate behavior so spinner/progress rendering does not race against stdout output.
+- Improve slides streaming/marker handling to avoid dropped interleaved text in chunked output.
+- Preserve deterministic text output in non-inline terminals (fallback path still prints slide paths in debug/non-inline contexts).
+
+### 4) Slide text fallback quality
+- Improve transcript fallback segmentation for slide blocks:
+  - reduce clipped starts
+  - better boundary handling for continuation fragments
+  - avoid over-aggressive replacement of long narrative bodies
+- Add regression coverage around truncated/ellipsis and chunk boundary behavior.
+
+## Why
+Users reported:
+- Needing cumbersome env remapping to use non-OpenAI-compatible providers.
+- `--model minimax`/provider aliases not being symmetrical/extensible.
+- Slides output truncation/misalignment and occasional paragraph clipping.
+- Kimi calls failing due to temperature constraints.
+
+This PR addresses those pain points directly.
+
+## User-facing behavior changes
+- New:
+  - `summarize "<url>" --model kimi`
+  - `summarize "<url>" --model kimi/<model-id>`
+- Existing workflows remain supported.
+- `--verbose` env diagnostics now include `kimiKey=true|false`.
+
+## Docs
+- `README.md`
+- `docs/llm.md`
+- `docs/config.md`
+- CLI help text (`src/run/help.ts`)
+
+## Tests
+Added/updated coverage for:
+- model spec parsing (`kimi`, `kimi/<model>`)
+- auto model candidate support for Kimi aliases
+- config env legacy key mapping (`apiKeys.kimi`)
+- daemon agent key selection for Kimi
+- temperature handling for Kimi
+- slides text/stream/progress regressions
+
+## Validation run
+```bash
+corepack pnpm exec vitest run \
+  tests/model-spec.test.ts \
+  tests/model-auto.test.ts \
+  tests/config.env.test.ts \
+  tests/daemon.agent.test.ts \
+  tests/llm.generate-text.test.ts \
+  tests/slides-output.stream-handler.test.ts \
+  tests/slides-text.utils.test.ts \
+  tests/progress-gate.test.ts
+
+corepack pnpm exec tsc -p tsconfig.build.json
+corepack pnpm exec node scripts/build-cli.mjs
+```
+
+## Notes / follow-up
+- There is a local scratch file `tmp-coerce-check.ts` in the working tree; this should not be included in the final PR unless intentionally needed.
+- This branch includes both provider and slides/progress work in one PR. If preferred, it can be split into:
+  1. provider/model support (Kimi/MiniMax-related)
+  2. slides/progress stability fixes

README.md

@@ -231,10 +231,14 @@ Notes:
 
 ### Model ids
 
-Use gateway-style ids: `<provider>/<model>`.
+Use gateway-style ids: `<provider>/<model>`. MiniMax and Kimi also support short aliases.
 
 Examples:
 
+- `minimax` (alias for MiniMax default model)
+- `minimax/minimax-m2.5`
+- `kimi` (alias for Kimi default model)
+- `kimi/kimi-k2.5`
 - `openai/gpt-5-mini`
 - `anthropic/claude-sonnet-4-5`
 - `xai/grok-4-fast-non-reasoning`
@@ -552,7 +556,11 @@ Set the key matching your chosen `--model`:
   - legacy `"apiKeys"` still works (mapped to env names)
 
 - `OPENAI_API_KEY` (for `openai/...`)
-- `NVIDIA_API_KEY` (for `nvidia/...`)
+- `NVIDIA_API_KEY` (for `nvidia/...`; supports `NGC_API_KEY` alias)
+- `MINIMAX_API_KEY` (for `minimax` and `minimax/...`)
+- `MINIMAX_BASE_URL` (optional MiniMax base URL override, default `https://api.minimax.io/v1`)
+- `KIMI_API_KEY` (for `kimi` and `kimi/...`; supports `MOONSHOT_API_KEY` alias)
+- `KIMI_BASE_URL` (optional Kimi base URL override, default `https://api.moonshot.ai/v1`; supports `MOONSHOT_BASE_URL` alias)
 - `ANTHROPIC_API_KEY` (for `anthropic/...`)
 - `XAI_API_KEY` (for `xai/...`)
 - `Z_AI_API_KEY` (for `zai/...`; supports `ZAI_API_KEY` alias)
@@ -636,6 +644,16 @@ Z.AI (OpenAI-compatible):
 - `Z_AI_API_KEY=...` (or `ZAI_API_KEY=...`)
 - Optional base URL override: `Z_AI_BASE_URL=...`
 
+MiniMax (OpenAI-compatible):
+
+- `MINIMAX_API_KEY=...`
+- Optional base URL override: `MINIMAX_BASE_URL=...`
+
+Kimi (OpenAI-compatible):
+
+- `KIMI_API_KEY=...` (or `MOONSHOT_API_KEY=...`)
+- Optional base URL override: `KIMI_BASE_URL=...` (or `MOONSHOT_BASE_URL=...`)
+
 Optional services:
 
 - `FIRECRAWL_API_KEY` (website extraction fallback)

docs/config.md

@@ -121,6 +121,8 @@ Legacy shortcut (still supported):
 {
   "apiKeys": {
     "openai": "sk-...",
+    "minimax": "...",
+    "kimi": "...",
     "anthropic": "sk-ant-...",
     "google": "...",
     "openrouter": "sk-or-...",

docs/llm.md

@@ -26,6 +26,10 @@ installed, auto mode can use local CLI models via `cli.enabled` or implicit auto
 - `OPENAI_USE_CHAT_COMPLETIONS` (optional; force OpenAI chat completions)
 - `NVIDIA_API_KEY` (required for `nvidia/...` models; alias: `NGC_API_KEY`)
 - `NVIDIA_BASE_URL` (optional; override NVIDIA OpenAI-compatible API endpoint; default: `https://integrate.api.nvidia.com/v1`)
+- `MINIMAX_API_KEY` (required for `minimax` and `minimax/...` models)
+- `MINIMAX_BASE_URL` (optional; override MiniMax base URL, default `https://api.minimax.io/v1`)
+- `KIMI_API_KEY` (required for `kimi` and `kimi/...` models; also accepts `MOONSHOT_API_KEY`)
+- `KIMI_BASE_URL` (optional; override Kimi base URL, default `https://api.moonshot.ai/v1`; also accepts `MOONSHOT_BASE_URL`)
 - `OPENROUTER_API_KEY` (optional; required for `openrouter/...` models; also used when `OPENAI_BASE_URL` points to OpenRouter)
 - `Z_AI_API_KEY` (required for `zai/...` models; supports `ZAI_API_KEY` alias)
 - `Z_AI_BASE_URL` (optional; override default Z.AI base URL)
@@ -47,6 +51,10 @@ installed, auto mode can use local CLI models via `cli.enabled` or implicit auto
     - `google/gemini-3-flash-preview`
     - `openai/gpt-5-mini`
     - `nvidia/z-ai/glm5`
+    - `minimax`
+    - `minimax/minimax-m2.5`
+    - `kimi`
+    - `kimi/kimi-k2.5`
     - `zai/glm-4.7`
     - `xai/grok-4-fast-non-reasoning`
     - `google/gemini-2.0-flash`
@@ -98,6 +106,16 @@ installed, auto mode can use local CLI models via `cli.enabled` or implicit auto
 
 Use `--model zai/<model>` (e.g. `zai/glm-4.7`). Defaults to Z.AI’s base URL and uses chat completions.
 
+## MiniMax
+
+Use `--model minimax` (default MiniMax model) or `--model minimax/<model>`.
+Requires `MINIMAX_API_KEY`. Requests use MiniMax's OpenAI-compatible endpoint by default.
+
+## Kimi
+
+Use `--model kimi` (default Kimi model) or `--model kimi/<model>`.
+Requires `KIMI_API_KEY` (or `MOONSHOT_API_KEY`). Requests use Kimi's OpenAI-compatible endpoint by default.
+
 ## Input limits
 
 - Text prompts are checked against the model’s max input tokens (LiteLLM catalog) using a GPT tokenizer.

packages/core/src/prompts/link-summary.ts

@@ -198,7 +198,26 @@ export function buildLinkSummaryPrompt({
   const listGuidanceLine =
     "Use short paragraphs; use bullet lists only when they improve scanability; avoid rigid templates.";
   const quoteGuidanceLine =
-    "Include 1-2 short exact excerpts (max 25 words each) formatted as Markdown italics using single asterisks when there is a strong, non-sponsor line. Use straight quotation marks (no curly) as needed. If no suitable line exists, omit excerpts. Never include ad/sponsor/boilerplate excerpts and do not mention them.";
+    slides && slides.count > 0
+      ? "For slide blocks, do not include direct quotes or transcript excerpts."
+      : "Include 1-2 short exact excerpts (max 25 words each) formatted as Markdown italics using single asterisks when there is a strong, non-sponsor line. Use straight quotation marks (no curly) as needed. If no suitable line exists, omit excerpts. Never include ad/sponsor/boilerplate excerpts and do not mention them.";
+  const slideSentenceGuidance = (() => {
+    if (!(slides && slides.count > 0)) return "";
+    if (preset === "short") return "For each slide block, write a concise neutral summary (2-3 sentences).";
+    if (preset === "medium") return "For each slide block, write a concise neutral summary (2-4 sentences).";
+    if (preset === "long") return "For each slide block, write a concise neutral summary (3-5 sentences).";
+    if (preset === "xl") return "For each slide block, write a concise neutral summary (3-6 sentences).";
+    return "For each slide block, write a concise neutral summary (4-7 sentences).";
+  })();
+  const slideToneInstruction =
+    slides && slides.count > 0
+      ? [
+          slideSentenceGuidance,
+          "Use third-person prose; avoid first-person/second-person voice and direct address.",
+          "Paraphrase transcript content; do not copy long conversational passages.",
+          "Exclude banter, filler phrases, and rhetorical asides.",
+        ].join("\n")
+      : "";
   const sponsorInstruction =
     hasTranscript || (slides && slides.count > 0)
       ? "Omit sponsor messages, ads, promos, and calls-to-action (including podcast ad reads), even if they appear in the transcript or slide timeline. Do not mention or acknowledge them, and do not say you skipped or ignored anything. Avoid sponsor/ad/promo language, brand names like Squarespace, or CTA phrases like discount code. Treat them as if they do not exist. If a slide segment is purely sponsor/ad content, leave that slide marker with no text."
@@ -222,9 +241,11 @@ export function buildLinkSummaryPrompt({
     "Format the answer in Markdown and obey the length-specific formatting above.",
     listGuidanceLine,
     quoteGuidanceLine,
+    slideToneInstruction,
     "Base everything strictly on the provided content and never invent details.",
     "Final check: remove any sponsor/ad references or mentions of skipping/ignoring content. Ensure excerpts (if any) are italicized and use only straight quotes.",
     'Final check for slides: every [slide:N] must be immediately followed by a line that starts with "## ". Remove any "Title:" or "Slide" label lines.',
+    "Final check for slides: rewrite transcript-like direct speech into concise neutral summaries.",
     timestampInstruction,
     shareGuidance,
     slideInstruction,

src/config.ts

@@ -88,6 +88,8 @@ export type NvidiaConfig = {
 export type ApiKeysConfig = {
   openai?: string;
   nvidia?: string;
+  minimax?: string;
+  kimi?: string;
   anthropic?: string;
   google?: string;
   xai?: string;
@@ -131,7 +133,7 @@ export type AutoRule = {
   /**
    * Candidate model ids (ordered).
    *
-   * - Native: `openai/...`, `google/...`, `xai/...`, `anthropic/...`, `zai/...`
+   * - Native: `openai/...`, `minimax`, `minimax/...`, `kimi`, `kimi/...`, `google/...`, `xai/...`, `anthropic/...`, `zai/...`
    * - OpenRouter (forced): `openrouter/<provider>/<model>` (e.g. `openrouter/openai/gpt-5-mini`)
    */
   candidates?: string[];
@@ -245,6 +247,8 @@ function resolveLegacyApiKeysEnv(apiKeys: ApiKeysConfig | undefined): EnvConfig
   const mapped: EnvConfig = {};
   if (typeof apiKeys.openai === "string") mapped.OPENAI_API_KEY = apiKeys.openai;
   if (typeof apiKeys.nvidia === "string") mapped.NVIDIA_API_KEY = apiKeys.nvidia;
+  if (typeof apiKeys.minimax === "string") mapped.MINIMAX_API_KEY = apiKeys.minimax;
+  if (typeof apiKeys.kimi === "string") mapped.KIMI_API_KEY = apiKeys.kimi;
   if (typeof apiKeys.anthropic === "string") mapped.ANTHROPIC_API_KEY = apiKeys.anthropic;
   if (typeof apiKeys.google === "string") mapped.GEMINI_API_KEY = apiKeys.google;
   if (typeof apiKeys.xai === "string") mapped.XAI_API_KEY = apiKeys.xai;
@@ -1194,6 +1198,8 @@ export function loadSummarizeConfig({ env }: { env: Record<string, string | unde
     const allowed = [
       "openai",
       "nvidia",
+      "minimax",
+      "kimi",
       "anthropic",
       "google",
       "xai",

src/daemon/agent.ts

@@ -396,12 +396,16 @@ type AgentApiKeys = {
   xaiApiKey: string | null;
   zaiApiKey: string | null;
   nvidiaApiKey: string | null;
+  minimaxApiKey: string | null;
+  kimiApiKey: string | null;
 };
 
 const REQUIRED_ENV_BY_PROVIDER: Record<string, string> = {
   openrouter: "OPENROUTER_API_KEY",
   openai: "OPENAI_API_KEY",
   nvidia: "NVIDIA_API_KEY",
+  minimax: "MINIMAX_API_KEY",
+  kimi: "KIMI_API_KEY",
   anthropic: "ANTHROPIC_API_KEY",
   google: "GEMINI_API_KEY",
   xai: "XAI_API_KEY",
@@ -423,6 +427,10 @@ function resolveApiKeyForModel({
         return apiKeys.openaiApiKey;
       case "nvidia":
         return apiKeys.nvidiaApiKey;
+      case "minimax":
+        return apiKeys.minimaxApiKey;
+      case "kimi":
+        return apiKeys.kimiApiKey;
       case "anthropic":
         return apiKeys.anthropicApiKey;
       case "google":
@@ -463,6 +471,10 @@ async function resolveAgentModel({
     googleApiKey,
     xaiApiKey,
     zaiApiKey,
+    minimaxApiKey,
+    minimaxBaseUrl,
+    kimiApiKey,
+    kimiBaseUrl,
     providerBaseUrls,
     zaiBaseUrl,
     nvidiaApiKey,
@@ -487,6 +499,8 @@ async function resolveAgentModel({
     xaiApiKey,
     zaiApiKey,
     nvidiaApiKey,
+    minimaxApiKey,
+    kimiApiKey,
   };
 
   const overrides = resolveRunOverrides({});
@@ -530,6 +544,39 @@ async function resolveAgentModel({
       return { ...resolved, maxOutputTokens, apiKeys };
     }
 
+    if (requestedModel.requiredEnv === "MINIMAX_API_KEY") {
+      const parsed = parseProviderModelId(requestedModel.llmModelId);
+      return {
+        provider: "openai",
+        model: resolveModelWithFallback({
+          provider: "openai",
+          modelId: parsed.model,
+          baseUrl: minimaxBaseUrl,
+        }),
+        maxOutputTokens,
+        apiKeys: {
+          ...apiKeys,
+          openaiApiKey: minimaxApiKey,
+        },
+      };
+    }
+    if (requestedModel.requiredEnv === "KIMI_API_KEY") {
+      const parsed = parseProviderModelId(requestedModel.llmModelId);
+      return {
+        provider: "openai",
+        model: resolveModelWithFallback({
+          provider: "openai",
+          modelId: parsed.model,
+          baseUrl: kimiBaseUrl,
+        }),
+        maxOutputTokens,
+        apiKeys: {
+          ...apiKeys,
+          openaiApiKey: kimiApiKey,
+        },
+      };
+    }
+
     const { provider, model } = parseProviderModelId(requestedModel.userModelId);
     const resolved = applyBaseUrlOverride(provider, model);
     return { ...resolved, maxOutputTokens, apiKeys };
@@ -555,6 +602,38 @@ async function resolveAgentModel({
   for (const attempt of attempts) {
     if (attempt.transport === "cli") continue;
     if (!envHasKey(envForAuto, attempt.requiredEnv)) continue;
+    if (attempt.requiredEnv === "MINIMAX_API_KEY") {
+      const parsed = parseProviderModelId(attempt.llmModelId ?? attempt.userModelId);
+      return {
+        provider: "openai",
+        model: resolveModelWithFallback({
+          provider: "openai",
+          modelId: parsed.model,
+          baseUrl: minimaxBaseUrl,
+        }),
+        maxOutputTokens,
+        apiKeys: {
+          ...apiKeys,
+          openaiApiKey: minimaxApiKey,
+        },
+      };
+    }
+    if (attempt.requiredEnv === "KIMI_API_KEY") {
+      const parsed = parseProviderModelId(attempt.llmModelId ?? attempt.userModelId);
+      return {
+        provider: "openai",
+        model: resolveModelWithFallback({
+          provider: "openai",
+          modelId: parsed.model,
+          baseUrl: kimiBaseUrl,
+        }),
+        maxOutputTokens,
+        apiKeys: {
+          ...apiKeys,
+          openaiApiKey: kimiApiKey,
+        },
+      };
+    }
     if (attempt.transport === "openrouter") {
       const modelId = attempt.userModelId.replace(/^openrouter\//i, "");
       const resolved = applyBaseUrlOverride("openrouter", modelId);