feat(compaction): Handle when models have input limit

Author: shantur

packages/opencode/src/provider/models.ts

@@ -48,6 +48,7 @@ export namespace ModelsDev {
     limit: z.object({
       context: z.number(),
       output: z.number(),
+      input: z.number().optional(),
     }),
     modalities: z
       .object({

packages/opencode/src/provider/provider.ts

@@ -465,6 +465,7 @@ export namespace Provider {
       limit: z.object({
         context: z.number(),
         output: z.number(),
+        input: z.number().optional(),
       }),
       status: z.enum(["alpha", "beta", "deprecated", "active"]),
       options: z.record(z.string(), z.any()),
@@ -527,6 +528,7 @@ export namespace Provider {
       limit: {
         context: model.limit.context,
         output: model.limit.output,
+        input: model.limit.input,
       },
       capabilities: {
         temperature: model.temperature,

packages/opencode/src/session/compaction.ts

@@ -7,13 +7,13 @@ import { Provider } from "../provider/provider"
 import { MessageV2 } from "./message-v2"
 import z from "zod"
 import { SessionPrompt } from "./prompt"
+import { Flag } from "../flag/flag"
 import { Token } from "../util/token"
 import { Log } from "../util/log"
 import { SessionProcessor } from "./processor"
 import { fn } from "@/util/fn"
 import { Agent } from "@/agent/agent"
 import { Plugin } from "@/plugin"
-import { Config } from "@/config/config"
 
 export namespace SessionCompaction {
   const log = Log.create({ service: "session.compaction" })
@@ -27,14 +27,15 @@ export namespace SessionCompaction {
     ),
   }
 
-  export async function isOverflow(input: { tokens: MessageV2.Assistant["tokens"]; model: Provider.Model }) {
-    const config = await Config.get()
-    if (config.compaction?.auto === false) return false
+  export const NEXT_INPUT_TOKEN_MAX = 20_000
+
+  export function isOverflow(input: { tokens: MessageV2.Assistant["tokens"]; model: Provider.Model }) {
+    if (Flag.OPENCODE_DISABLE_AUTOCOMPACT) 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 usable = context - output
+    const usable = input.model.limit.input ? input.model.limit.input - NEXT_INPUT_TOKEN_MAX : context - output
     return count > usable
   }
 
@@ -47,8 +48,7 @@ export namespace SessionCompaction {
   // calls. then erases output of previous tool calls. idea is to throw away old
   // tool calls that are no longer relevant.
   export async function prune(input: { sessionID: string }) {
-    const config = await Config.get()
-    if (config.compaction?.prune === false) return
+    if (Flag.OPENCODE_DISABLE_PRUNE) return
     log.info("pruning")
     const msgs = await Session.messages({ sessionID: input.sessionID })
     let total = 0