diff --git a/packages/core/src/config/compaction.ts b/packages/core/src/config/compaction.ts
index 3c5960c83556..3a9df791c615 100644
--- a/packages/core/src/config/compaction.ts
+++ b/packages/core/src/config/compaction.ts
@@ -12,4 +12,5 @@ export class Info extends Schema.Class<Info>("ConfigV2.Compaction")({
   prune: Schema.Boolean.pipe(Schema.optional),
   keep: Keep.pipe(Schema.optional),
   buffer: NonNegativeInt.pipe(Schema.optional),
+  max_request_bytes: NonNegativeInt.pipe(Schema.optional),
 }) {}
diff --git a/packages/core/src/session/compaction.ts b/packages/core/src/session/compaction.ts
index ba52eb77dd00..5b0a245f0a1d 100644
--- a/packages/core/src/session/compaction.ts
+++ b/packages/core/src/session/compaction.ts
@@ -47,6 +47,7 @@ type Settings = {
   readonly auto: boolean
   readonly buffer: number
   readonly tokens: number
+  readonly maxRequestBytes?: number
 }
 
 type Dependencies = {
@@ -64,7 +65,9 @@ type Input = {
   readonly request: LLMRequest
 }
 
-const estimate = (value: unknown) => Token.estimate(JSON.stringify(value))
+const stringify = (value: unknown) => JSON.stringify(value)
+const estimate = (value: unknown) => Token.estimate(stringify(value))
+const byteLength = (value: string) => new TextEncoder().encode(value).length
 
 const truncate = (value: string) =>
   value.length <= TOOL_OUTPUT_MAX_CHARS ? value : `${value.slice(0, TOOL_OUTPUT_MAX_CHARS)}\n[truncated]`
@@ -113,6 +116,7 @@ const settings = (documents: readonly Config.Entry[]) => {
       auto: current.auto ?? result.auto,
       buffer: current.buffer ?? result.buffer,
       tokens: current.keep?.tokens ?? result.tokens,
+      maxRequestBytes: current.max_request_bytes ?? result.maxRequestBytes,
     }),
     { auto: true, buffer: DEFAULT_BUFFER, tokens: DEFAULT_KEEP_TOKENS },
   )
@@ -220,9 +224,14 @@ export const make = (dependencies: Dependencies) => {
     const context = input.model.route.defaults.limits?.context
     if (context === undefined || context <= 0) return false
     const output = input.request.generation?.maxTokens ?? input.model.route.defaults.limits?.output ?? 0
+    const payload = stringify({
+      system: input.request.system,
+      messages: input.request.messages,
+      tools: input.request.tools,
+    })
     if (
-      estimate({ system: input.request.system, messages: input.request.messages, tools: input.request.tools }) <=
-      context - Math.max(output, config.buffer)
+      Token.estimate(payload) <= context - Math.max(output, config.buffer) &&
+      (config.maxRequestBytes === undefined || byteLength(payload) <= config.maxRequestBytes)
     )
       return false
     return yield* compactAfterOverflow(input)
diff --git a/packages/core/test/config/config.test.ts b/packages/core/test/config/config.test.ts
index c3c42cab3022..419450f1dfba 100644
--- a/packages/core/test/config/config.test.ts
+++ b/packages/core/test/config/config.test.ts
@@ -340,6 +340,7 @@ describe("Config", () => {
                   prune: false,
                   keep: { tokens: 2000 },
                   buffer: 10000,
+                  max_request_bytes: 1048576,
                 },
                 skills: ["./skills", "~/shared-skills", "https://example.com/.well-known/skills/"],
                 instructions: ["CONTRIBUTING.md", ".cursor/rules/*.md", "https://example.com/shared-rules.md"],
@@ -426,6 +427,7 @@ describe("Config", () => {
               prune: false,
               keep: { tokens: 2000 },
               buffer: 10000,
+              max_request_bytes: 1048576,
             })
             expect(documents[0]?.info.skills).toEqual([
               "./skills",
diff --git a/packages/core/test/session-compaction.test.ts b/packages/core/test/session-compaction.test.ts
index e91c89c00954..8d087b781014 100644
--- a/packages/core/test/session-compaction.test.ts
+++ b/packages/core/test/session-compaction.test.ts
@@ -1,5 +1,13 @@
 import { expect, test } from "bun:test"
+import { Config } from "@opencode-ai/core/config"
+import { ConfigCompaction } from "@opencode-ai/core/config/compaction"
+import { EventV2 } from "@opencode-ai/core/event"
 import { SessionCompaction } from "@opencode-ai/core/session/compaction"
+import { SessionMessage } from "@opencode-ai/core/session/message"
+import { SessionSchema } from "@opencode-ai/core/session/schema"
+import { LLM, LLMEvent, Message, Model } from "@opencode-ai/llm"
+import { route } from "@opencode-ai/llm/protocols/openai-chat"
+import { DateTime, Effect, Stream } from "effect"
 
 test("compaction describes tool media without embedding base64", () => {
   const base64 = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAAB"
@@ -16,3 +24,78 @@ test("compaction describes tool media without embedding base64", () => {
   expect(serialized).toBe("Image read successfully\n[Attached image/png: pixel.png]")
   expect(serialized).not.toContain(base64)
 })
+
+test("compaction can trigger from configured request byte envelope", async () => {
+  const events: Array<{ readonly type: string; readonly payload: unknown }> = []
+  const compact = SessionCompaction.make({
+    config: [
+      new Config.Document({
+        type: "document",
+        info: new Config.Info({
+          compaction: new ConfigCompaction.Info({
+            keep: new ConfigCompaction.Keep({ tokens: 16 }),
+            max_request_bytes: 512,
+          }),
+        }),
+      }),
+    ],
+    events: {
+      publish: (definition, payload) =>
+        Effect.sync(() => {
+          const event = { id: EventV2.ID.create(), type: definition.type, data: payload } as EventV2.Payload<
+            typeof definition
+          >
+          events.push({ type: definition.type, payload })
+          return event
+        }),
+      subscribe: () => Stream.empty,
+      all: () => Stream.empty,
+      durable: () => Stream.empty,
+      listen: () => Effect.succeed(Effect.void),
+      project: () => Effect.void,
+      replay: () => Effect.void,
+      replayAll: () => Effect.succeed(undefined),
+      remove: () => Effect.void,
+      claim: () => Effect.void,
+    },
+    llm: {
+      stream: () =>
+        Stream.make(
+          LLMEvent.textStart({ id: "summary" }),
+          LLMEvent.textDelta({ id: "summary", text: "## Objective\n- Keep request small" }),
+          LLMEvent.finish({ reason: "stop" }),
+        ),
+    },
+  })
+  const sessionID = SessionSchema.ID.make("ses_byte_guard")
+  const model = Model.make({
+    id: "claude-fable-5",
+    provider: "opencode",
+    route: route.with({ limits: { context: 1_000_000, output: 4_096 } }),
+  })
+  const message: SessionMessage.Message = {
+    id: SessionMessage.ID.create(),
+    type: "user",
+    text: "Important historical context ".repeat(80),
+    files: [],
+    agents: [],
+    time: { created: DateTime.makeUnsafe(Date.now()) },
+  }
+
+  const result = await Effect.runPromise(
+    compact.compactIfNeeded({
+      sessionID,
+      entries: [{ seq: 1, message }],
+      model,
+      request: LLM.request({
+        model,
+        messages: [Message.user("x".repeat(1_500))],
+        tools: [],
+      }),
+    }),
+  )
+
+  expect(result).toBe(true)
+  expect(events).toHaveLength(2)
+  expect(events.map((event) => (event.payload as { readonly reason?: string }).reason)).toEqual(["auto", "auto"])
+})