Declare modalities so image input actually reaches the model

src/index.ts

@@ -69,21 +69,35 @@ const MODEL_TTL_MS = 6 * 60 * 60 * 1000; // 6h
 
 // Static baseline (context windows from the CLI). Seeds the provider on first
 // run / when logged out; live discovery replaces it once a request has run.
+// Keep the capability flags in sync with what /models reports — a baseline that
+// understates them registers the model as text-only until the cache warms.
 const STATIC_MODELS: DiscoveredModel[] = [
-  { id: "k3", display_name: "Kimi K3", context_length: 1_048_576, supports_reasoning: true, supports_tool_use: true },
+  {
+    id: "k3",
+    display_name: "Kimi K3",
+    context_length: 1_048_576,
+    supports_reasoning: true,
+    supports_tool_use: true,
+    supports_image_in: true,
+    supports_video_in: true,
+  },
   {
     id: "kimi-for-coding",
     display_name: "Kimi K2.7 Coding",
     context_length: 262_144,
     supports_reasoning: true,
     supports_tool_use: true,
+    supports_image_in: true,
+    supports_video_in: true,
   },
   {
     id: "kimi-for-coding-highspeed",
     display_name: "Kimi K2.7 Coding Highspeed",
     context_length: 262_144,
     supports_reasoning: true,
     supports_tool_use: true,
+    supports_image_in: true,
+    supports_video_in: true,
   },
 ];
 

src/internal.ts

@@ -21,6 +21,9 @@ export type UpstreamModel = {
 // /models fetch filter).
 export type DiscoveredModel = UpstreamModel & { id: string };
 
+// OpenCode's model-config modality literals.
+export type Modality = "text" | "audio" | "image" | "video" | "pdf";
+
 export type OAuthRecord = { type: "oauth"; access: string; refresh: string; expires: number };
 
 export type ModelCache = { ts: number; models: DiscoveredModel[] };
@@ -59,12 +62,21 @@ export function toAuth(data: Record<string, any>, now = Date.now()): OAuthRecord
 
 // Map an upstream model to OpenCode's provider-config model shape.
 export function toConfigModel(m: DiscoveredModel) {
+  const input: Modality[] = ["text"];
+  if (m.supports_image_in) input.push("image");
+  if (m.supports_video_in) input.push("video");
+
   return {
     name: m.display_name || m.id,
     limit: { context: Number(m.context_length) || 262_144, output: DEFAULT_OUTPUT_LIMIT },
     reasoning: !!m.supports_reasoning || m.supports_thinking_type === "only" || m.supports_thinking_type === "both",
     tool_call: m.supports_tool_use !== false,
     attachment: !!(m.supports_image_in || m.supports_video_in),
+    // `attachment` alone is not enough: OpenCode gates what it will actually
+    // send on `modalities.input`, which defaults to text-only. Without this an
+    // image-capable model still has its attachments dropped before the request
+    // is built, and it replies that it cannot read images.
+    modalities: { input, output: ["text"] as Modality[] },
     // kimi coding models only accept temperature === 1; hide the knob.
     temperature: false,
   };

test/unit.test.mjs

@@ -77,6 +77,21 @@ describe("toConfigModel", () => {
     assert.equal(toConfigModel({ id: "m", supports_video_in: true }).attachment, true);
     assert.equal(toConfigModel({ id: "m" }).attachment, false);
   });
+  it("mirrors input support into modalities so OpenCode actually sends them", () => {
+    assert.deepEqual(toConfigModel({ id: "m", supports_image_in: true }).modalities.input, ["text", "image"]);
+    assert.deepEqual(toConfigModel({ id: "m", supports_video_in: true }).modalities.input, ["text", "video"]);
+    assert.deepEqual(toConfigModel({ id: "m", supports_image_in: true, supports_video_in: true }).modalities.input, [
+      "text",
+      "image",
+      "video",
+    ]);
+  });
+  it("keeps modalities text-only when the model has no image/video input", () => {
+    assert.deepEqual(toConfigModel({ id: "m" }).modalities.input, ["text"]);
+  });
+  it("always reports text output", () => {
+    assert.deepEqual(toConfigModel({ id: "m", supports_image_in: true }).modalities.output, ["text"]);
+  });
   it("always hides the temperature knob", () => {
     assert.equal(toConfigModel({ id: "m" }).temperature, false);
   });