diff --git a/packages/api/src/agents/__tests__/run-summarization.test.ts b/packages/api/src/agents/__tests__/run-summarization.test.ts
index bfb9ab13cd..6d3e923f23 100644
--- a/packages/api/src/agents/__tests__/run-summarization.test.ts
+++ b/packages/api/src/agents/__tests__/run-summarization.test.ts
@@ -648,6 +648,35 @@ describe('summarizationEnabled resolution', () => {
     expect(config.provider).toBe('openAI');
     expect(config.model).toBe('gpt-4o');
   });
+
+  it('false when the effective context budget is below the viable minimum', async () => {
+    /**
+     * A tiny user-set maxContextTokens re-triggers summarization on every
+     * graph step until the recursion limit aborts the run; the guard falls
+     * back to plain pruning instead.
+     */
+    const agents = await callAndCapture({
+      agents: [makeAgent({ maxContextTokens: 10 })],
+      summarizationConfig: {
+        enabled: true,
+        provider: 'anthropic',
+        model: 'claude-3-haiku',
+      },
+    });
+    expect(agents[0].summarizationEnabled).toBe(false);
+  });
+
+  it('true at exactly the 1024-token viable minimum', async () => {
+    const agents = await callAndCapture({
+      agents: [makeAgent({ maxContextTokens: 1024 })],
+      summarizationConfig: {
+        enabled: true,
+        provider: 'anthropic',
+        model: 'claude-3-haiku',
+      },
+    });
+    expect(agents[0].summarizationEnabled).toBe(true);
+  });
 });
 
 // ---------------------------------------------------------------------------
diff --git a/packages/api/src/agents/run.ts b/packages/api/src/agents/run.ts
index 77cff7a4e7..648d71d572 100644
--- a/packages/api/src/agents/run.ts
+++ b/packages/api/src/agents/run.ts
@@ -941,6 +941,18 @@ function shapeSummarizationConfig(
   };
 }
 
+/**
+ * Below this context budget a summarization cycle cannot make progress: the
+ * summary allocation rounds down to a handful of tokens, the rewritten history
+ * still overflows, and the graph re-triggers summarization on every step until
+ * the recursion limit aborts the run — dozens of wasted LLM calls surfaced to
+ * the user as an opaque LangGraph error. Falling back to plain pruning instead
+ * either fits the request or fails fast with the actionable `empty_messages`
+ * token-budget breakdown. Matches the floor `initializeAgent` applies when the
+ * user supplies no override.
+ */
+const MIN_SUMMARIZATION_CONTEXT_TOKENS = 1024;
+
 /**
  * Applies `reserveRatio` against the pre-ratio base context budget, falling
  * back to the pre-computed `maxContextTokens` from initializeAgent.
@@ -1879,6 +1891,20 @@ export async function createRun({
       agent.maxContextTokens,
     );
 
+    const summarizationViable =
+      effectiveMaxContextTokens == null ||
+      effectiveMaxContextTokens >= MIN_SUMMARIZATION_CONTEXT_TOKENS;
+    if (summarization.enabled && !summarizationViable) {
+      logger.warn(
+        '[createRun] Summarization disabled for this run: context budget below viable minimum',
+        {
+          agentId: agent.id,
+          effectiveMaxContextTokens,
+          minimum: MIN_SUMMARIZATION_CONTEXT_TOKENS,
+        },
+      );
+    }
+
     const reasoningKey = getReasoningKey(provider, llmConfig, agent.endpoint, agent.reasoningKey);
     const agentInput: AgentInputs = {
       provider,
@@ -1896,7 +1922,7 @@ export async function createRun({
       useLegacyContent: agent.useLegacyContent ?? false,
       discoveredTools:
         !isSubagent && discoveredTools.size > 0 ? Array.from(discoveredTools) : undefined,
-      summarizationEnabled: summarization.enabled,
+      summarizationEnabled: summarization.enabled && summarizationViable,
       summarizationConfig: summarization.config,
       ...(!isSubagent && compactionSemanticIndex != null ? { compactionSemanticIndex } : {}),
       initialSummary: isSubagent ? undefined : initialSummary,