opencode-see-image 0.8.6 → 0.9.1

package/.claude/settings.local.json

@@ -0,0 +1,12 @@
+{
+  "permissions": {
+    "allow": [
+      "Bash(bun build *)",
+      "Bash(npm view *)",
+      "Bash(FILTER_BRANCH_SQUELCH_WARNING=1 git filter-branch -f --msg-filter 'sed \"/Co-Authored-By: Claude/d\" | sed -e :a -e \"/^\\\\n*$/{\\\\$d;N;ba\" -e \"}\"' HEAD~2..HEAD)",
+      "Bash(echo \"--- created, exit $? ---\")",
+      "Bash(node -p \"require\\('./package.json'\\).version\")",
+      "Bash(echo \"local package.json version: $\\(node -p \"require\\('./package.json'\\).version\" \\)\")"
+    ]
+  }
+}

package/README.md

@@ -42,7 +42,7 @@ You need a connected vision-capable provider. The plugin auto-detects whichever
 2. Select **opencode** (OpenCode Zen)
 3. Paste your API key from [opencode.ai/auth](https://opencode.ai/auth)
 
-The plugin falls back to **big-pickle** (~12000ms). No subscription needed.
+The plugin falls back to **mimo-v2.5-free**. No subscription needed.
 
 ### Paid, w/ OpenCode Go
 1. Run `/connect` in opencode
@@ -55,7 +55,7 @@ The plugin prefers **minimax-m3** via opencode-go (~3000ms) when available.
 
 Set the `SEE_IMAGE_*` env vars to point at any Anthropic-Messages-compatible endpoint. See [Configuration](#configuration) below.
 
-**Resolution order:** explicit `SEE_IMAGE_API_KEY` env → configured `SEE_IMAGE_PROVIDER` → `opencode-go` (MiniMax M3) → `opencode` (big-pickle, free).
+**Resolution order:** explicit `SEE_IMAGE_API_KEY` env → configured `SEE_IMAGE_PROVIDER` → `opencode-go` (MiniMax M3) → `opencode` (mimo-v2.5-free, free).
 
 ## How it works
 
@@ -126,7 +126,8 @@ export SEE_IMAGE_MODEL="kimi-k2.7-code"
 
 | Model | Speed | Notes |
 |---|---|---|
-| `big-pickle` | ~12000ms | Free. Accurate. Default fallback when only Zen is connected. |
+| `mimo-v2.5-free` | — | Free. Default fallback when only Zen is connected (routed via CLI). |
+| `big-pickle` | ~12000ms | Free. Accurate. Alternative Zen fallback. |
 
 **Paid (OpenCode Go):**
 

package/index.ts

@@ -11,7 +11,7 @@ const ENDPOINT =
   "https://opencode.ai/zen/go/v1/messages"
 const MODEL = process.env.SEE_IMAGE_MODEL || "minimax-m3"
 const PROVIDER_ID = process.env.SEE_IMAGE_PROVIDER || "opencode-go"
-const TIMEOUT = parseInt(process.env.SEE_IMAGE_TIMEOUT || "10000", 10)
+const TIMEOUT = parseInt(process.env.SEE_IMAGE_TIMEOUT || "30000", 10)
 const API_VERSION = process.env.SEE_IMAGE_API_VERSION || "2023-06-01"
 const USER_AGENT =
   process.env.SEE_IMAGE_USER_AGENT ||
@@ -47,8 +47,9 @@ function resolveFromDb(
   const dbPath = opencodeDbPath()
   if (!fs.existsSync(dbPath)) return null
 
+  let db: Database | undefined
   try {
-    const db = new Database(dbPath, { readonly: true })
+    db = new Database(dbPath, { readonly: true })
     let rows: Array<{ data: string }>
 
     if (!filename || filename === "clipboard") {
@@ -98,8 +99,6 @@ function resolveFromDb(
       }
     }
 
-    db.close()
-
     if (!rows.length) return null
     const part = JSON.parse(rows[0].data)
     const url: string = part.url || ""
@@ -112,6 +111,8 @@ function resolveFromDb(
     }
   } catch {
     return null
+  } finally {
+    db?.close()
   }
 }
 
@@ -217,6 +218,7 @@ function readProviderKey(providerID: string): string | null {
 
 async function seeImageViaSDK(
   client: any,
+  $: any,
   dataUrl: string,
   mediaType: string,
   prompt: string,
@@ -224,14 +226,40 @@ async function seeImageViaSDK(
 ): Promise<{ text: string; model: string; provider: string }> {
   const errors: string[] = []
 
-  // Write image to a temp file so the server can read it directly
+  // Write image to a temp file so the server can read it directly. Use the
+  // real extension so the CLI can sniff the type correctly.
   const b64 = dataUrl.split(",")[1] || ""
-  const tmpPath = path.join(os.tmpdir(), `see-image-${Date.now()}.png`)
+  const ext =
+    Object.entries(EXT_MEDIA).find(([, m]) => m === mediaType)?.[0] || "png"
+  const tmpPath = path.join(os.tmpdir(), `see-image-${Date.now()}.${ext}`)
   try {
     fs.writeFileSync(tmpPath, Buffer.from(b64, "base64"))
   } catch {}
 
-  const fileUrl = tmpPath
+  // For free opencode models, use CLI instead of SDK (SDK returns empty).
+  // Bun's $ doesn't accept an AbortSignal, so race the output against a
+  // timeout to actually bound how long a slow model can hang us.
+  const freeFallback = async (modelID: string, userPrompt: string): Promise<string | null> => {
+    try {
+      const proc = $`opencode run -f ${tmpPath} -m opencode/${modelID} ${userPrompt} --format json --dangerously-skip-permissions`.nothrow()
+      const out = await Promise.race([
+        proc.text(),
+        new Promise<never>((_, reject) =>
+          setTimeout(() => reject(new Error(`timed out after ${TIMEOUT}ms`)), TIMEOUT),
+        ),
+      ])
+      for (const line of out.split("\n").filter(Boolean)) {
+        try {
+          const parsed = JSON.parse(line)
+          if (parsed?.part?.type === "text" && parsed?.part?.text) {
+            return parsed.part.text
+          }
+        } catch {}
+      }
+    } catch {}
+    return null
+  }
+
   let result: { text: string; model: string; provider: string } | undefined
 
   try {
@@ -245,6 +273,17 @@ async function seeImageViaSDK(
     candidates.push({ providerID: "opencode", modelID: "mimo-v2.5-free" })
 
     for (const { providerID, modelID } of candidates) {
+      if (providerID === "opencode") {
+        // SDK session.prompt returns empty for free models; use CLI instead
+        const text = await freeFallback(modelID, prompt)
+        if (text) {
+          result = { text, model: modelID, provider: providerID }
+          break
+        }
+        errors.push(`${providerID}/${modelID}: no text from CLI fallback`)
+        continue
+      }
+
       let sessionID: string | undefined
       try {
         const sessionRes = await client.session.create({ body: {} })
@@ -255,22 +294,29 @@ async function seeImageViaSDK(
         }
 
         const controller = new AbortController()
+        const onAbort = () => controller.abort()
+        abort?.addEventListener("abort", onAbort)
         const timer = setTimeout(() => controller.abort(), TIMEOUT)
-        const res = await client.session.prompt({
-          path: { id: sessionID },
-          body: {
-            model: { providerID, modelID },
-            parts: [
-              { type: "file", mime: mediaType, url: providerID === "opencode" ? fileUrl : dataUrl },
-              { type: "text", text: prompt },
-            ],
-            tools: {},
-            system:
-              "You are a vision assistant. Describe the image accurately and concisely. Answer with text only.",
-          },
-          signal: controller.signal,
-        })
-        clearTimeout(timer)
+        let res
+        try {
+          res = await client.session.prompt({
+            path: { id: sessionID },
+            body: {
+              model: { providerID, modelID },
+              parts: [
+                { type: "file", mime: mediaType, url: dataUrl },
+                { type: "text", text: prompt },
+              ],
+              tools: {},
+              system:
+                "You are a vision assistant. Describe the image accurately and concisely. Answer with text only.",
+            },
+            signal: controller.signal,
+          })
+        } finally {
+          clearTimeout(timer)
+          abort?.removeEventListener("abort", onAbort)
+        }
 
         const parts = res.data?.parts ?? []
         const text = (parts as any[])
@@ -312,7 +358,7 @@ async function seeImageViaSDK(
 
     const errMsg = errors.join("; ")
     const hint = errMsg.includes("usage limit")
-      ? ` Enable usage from your balance at https://opencode.ai/workspace/wrk_01KVARG0A0Y87XV5JYBNJ0WRXB/go`
+      ? ` Enable usage from your balance in your opencode workspace at https://opencode.ai/workspace`
       : ""
     throw new Error(
       `see_image: SDK vision call failed for all candidates. ${errMsg}.${hint}`,
@@ -454,6 +500,7 @@ const SeeImagePlugin: Plugin = async (ctx) => {
       } else {
         result = await seeImageViaSDK(
           client,
+          $,
           resolved.dataUrl,
           resolved.mediaType,
           prompt,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-see-image",
-  "version": "0.8.6",
+  "version": "0.9.1",
   "description": "Give non-vision opencode models the ability to see images/screenshots by routing them to a vision-capable model (MiniMax M3 via opencode-go by default).",
   "type": "module",
   "main": "index.ts",