From bb7056b905ae7a14af1e7c22d18d22ba88bbed08 Mon Sep 17 00:00:00 2001
From: Felipe Madero <felipe.madero@gmail.com>
Date: Mon, 26 Jan 2026 23:06:49 -0300
Subject: [PATCH 1/2] feat(provider): auto-detect Ollama context limits

Query Ollama API to get model context limits (num_ctx) for proper
context percentage display in status bar. Detects Ollama servers
by checking if root endpoint returns "Ollama is running", then
fetches model info via /api/show. Falls back to 4096 default.
---
 packages/opencode/src/provider/provider.ts | 47 ++++++++++++++++++++++
 1 file changed, 47 insertions(+)

diff --git a/packages/opencode/src/provider/provider.ts b/packages/opencode/src/provider/provider.ts
index ee7ee75c9f5..28bca25a272 100644
--- a/packages/opencode/src/provider/provider.ts
+++ b/packages/opencode/src/provider/provider.ts
@@ -41,6 +41,41 @@ import { ProviderTransform } from "./transform"
 export namespace Provider {
   const log = Log.create({ service: "provider" })
 
+  const DEFAULT_OLLAMA_CONTEXT = 4096
+
+  async function isOllamaServer(baseURL: string): Promise<boolean> {
+    try {
+      const ollamaBase = baseURL.replace(/\/v1\/?$/, "")
+      const response = await fetch(ollamaBase, {
+        signal: AbortSignal.timeout(2000),
+      })
+      if (!response.ok) return false
+      const text = await response.text()
+      return text === "Ollama is running"
+    } catch {
+      return false
+    }
+  }
+
+  async function fetchOllamaModelContext(baseURL: string, modelName: string): Promise<number> {
+    try {
+      const ollamaBase = baseURL.replace(/\/v1\/?$/, "")
+      const response = await fetch(`${ollamaBase}/api/show`, {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ name: modelName }),
+        signal: AbortSignal.timeout(2000),
+      })
+      if (!response.ok) return DEFAULT_OLLAMA_CONTEXT
+      const data = (await response.json()) as { parameters?: string }
+      const match = data.parameters?.match(/num_ctx\s+(\d+)/)
+      if (match) return parseInt(match[1], 10)
+    } catch {
+      // Error querying Ollama - use default
+    }
+    return DEFAULT_OLLAMA_CONTEXT
+  }
+
   function isGpt5OrLater(modelID: string): boolean {
     const match = /^gpt-(\d+)/.exec(modelID)
     if (!match) {
@@ -806,6 +841,18 @@ export namespace Provider {
         parsed.models[modelID] = parsedModel
       }
       database[providerID] = parsed
+
+      // Fetch context limits from Ollama API if this is an Ollama server
+      // Only query Ollama if no limit was configured (config takes priority)
+      if (parsed.options.baseURL && (await isOllamaServer(parsed.options.baseURL))) {
+        const contextFetches = Object.entries(parsed.models).map(async ([modelID, model]) => {
+          if (model.limit.context === 0) {
+            const context = await fetchOllamaModelContext(parsed.options.baseURL, modelID)
+            model.limit.context = context
+          }
+        })
+        await Promise.all(contextFetches)
+      }
     }
 
     // load env

From 10aa64bf982017a9820f12085a389ade3bd720d7 Mon Sep 17 00:00:00 2001
From: Felipe Madero <felipe.madero@gmail.com>
Date: Mon, 26 Jan 2026 23:51:22 -0300
Subject: [PATCH 2/2] fix(compaction): handle models without output limit

When limit.output is 0, fall back to reserving 20% of context
(capped at OUTPUT_TOKEN_MAX) instead of hardcoded 32000.
This fixes compaction triggering immediately on small context
models like 16k Ollama models.
---
 packages/opencode/src/session/compaction.ts | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/packages/opencode/src/session/compaction.ts b/packages/opencode/src/session/compaction.ts
index fb382530291..8f9044f2c06 100644
--- a/packages/opencode/src/session/compaction.ts
+++ b/packages/opencode/src/session/compaction.ts
@@ -27,13 +27,15 @@ export namespace SessionCompaction {
     ),
   }
 
+  const OUTPUT_RESERVE_RATIO = 0.10
+
   export async function isOverflow(input: { tokens: MessageV2.Assistant["tokens"]; model: Provider.Model }) {
     const config = await Config.get()
     if (config.compaction?.auto === false) return false
     const context = input.model.limit.context
     if (context === 0) return false
     const count = input.tokens.input + input.tokens.cache.read + input.tokens.output
-    const output = Math.min(input.model.limit.output, SessionPrompt.OUTPUT_TOKEN_MAX) || SessionPrompt.OUTPUT_TOKEN_MAX
+    const output = Math.min(input.model.limit.output, SessionPrompt.OUTPUT_TOKEN_MAX) || Math.min(Math.floor(context * OUTPUT_RESERVE_RATIO), SessionPrompt.OUTPUT_TOKEN_MAX)
     const usable = input.model.limit.input || context - output
     return count > usable
   }