@bd7pil/opencode-deep-memory 0.8.0 → 0.8.2

package/README.md

@@ -177,58 +177,38 @@ npm run verify   # typecheck + test (363) + build + smoke (49)
 
 ## Acknowledgments
 
-**[MiMo-Code][]** — a terminal-native AI coding assistant with persistent memory that keeps a
-deep understanding of your project across sessions while continuously improving itself.
+**[DCP][]** — Dynamic Context Pruning for OpenCode. Tool dedup, error purge, and nudge system.
 
-**[Magic Context][]** — unbounded context. Memory that manages itself. One session, for life.
-The hippocampus for coding agents, part of CortexKit.
+**[Headroom][]** — JSON array crush and CCR (Compress-Cache-Retrieve).
 
-**[Aider][]** — AI pair programming in your terminal. Lets you pair program with LLMs to start
-a new project or build on your existing codebase.
+**[Edgee][]** — Per-tool compression strategies (read, bash, grep, glob).
 
-**[Roo Code][]** — a whole dev team of AI agents in your code editor.
+**[Contextomizer][]** — Content type detection pipeline.
 
-**[Continue][]** — pioneering open-source coding agent, available as a CLI, VS Code extension,
-and JetBrains plugin.
+**[Focus Agent][]** — Absolute token thresholds and assistant text compression research.
 
-**[OpenHands][]** — Code Less, Make More. A community focused on AI-driven development.
+**[LLMLingua][]** — Selective compression: preserve structure, compress prose.
 
-**[Plandex][]** — an AI coding agent designed for large tasks and real world projects.
+**[Codex CLI][]** — Handoff prefix pattern for compaction continuity.
 
-**[DCP][]** — Dynamic Context Pruning for OpenCode. Our tool deduplication, error purging,
-and nudge system are inspired by DCP's architecture.
+**[Google ADK][]** — Append-only event compaction architecture.
 
-**[Headroom][]** — compress tool outputs, logs, files, RAG chunks for AI agents.
-Our JSON array crush and CCR (Compress-Cache-Retrieve) are derived from Headroom's SmartCrusher.
+**[Hermes][]** — 8-section structured compaction prompt design.
 
-**[Edgee][]** — agent gateway that compresses tokens before LLM providers.
-Our per-tool compression strategies (read, bash, grep, glob) are inspired by Edgee's approach.
+**[MiMo-Code][]** — Terminal-native AI coding assistant with persistent memory.
 
-**[Contextomizer][]** — ultra-fast deterministic library for transforming bloated tool outputs.
-Our content type detection pipeline is inspired by Contextomizer's approach.
+**[Magic Context][]** — Unbounded context for coding agents.
 
-**[Focus Agent][]** — autonomous memory management for coding agents.
-Our absolute token thresholds and assistant text compression strategy are based on Focus Agent's research.
+**[Aider][]** — AI pair programming in your terminal.
 
-**[LLMLingua][]** — prompt compression for LLMs.
-Our selective assistant text compression (preserve structure, compress prose) is inspired by LLMLingua's approach.
+**[Roo Code][]** — A whole dev team of AI agents in your code editor.
 
-**[Codex CLI][]** — OpenAI's coding agent.
-Our handoff prefix pattern (telling the LLM it's resuming a prior task) is based on Codex CLI's compaction protocol.
+**[Continue][]** — Pioneering open-source coding agent.
 
-**[Google ADK][]** — Agent Development Kit with append-only event compaction.
-Our structured compaction prompt (Hermes-8 sections) is inspired by ADK's compaction architecture.
+**[OpenHands][]** — Code Less, Make More.
 
-**[Hermes][]** — production-grade compaction prompt design.
-Our 8-section checkpoint template follows Hermes's structured summary format.
+**[Plandex][]** — AI coding agent for large tasks and real world projects.
 
-[MiMo-Code]: https://github.com/XiaomiMiMo/MiMo-Code
-[Magic Context]: https://github.com/cortexkit/magic-context
-[Aider]: https://github.com/Aider-AI/aider
-[Roo Code]: https://github.com/RooCodeInc/Roo-Code
-[Continue]: https://github.com/continuedev/continue
-[OpenHands]: https://github.com/All-Hands-AI/OpenHands
-[Plandex]: https://github.com/plandex-ai/plandex
 [DCP]: https://github.com/Opencode-DCP/opencode-dynamic-context-pruning
 [Headroom]: https://github.com/chopratejas/headroom
 [Edgee]: https://github.com/edgee-ai/edgee
@@ -238,6 +218,13 @@ Our 8-section checkpoint template follows Hermes's structured summary format.
 [Codex CLI]: https://github.com/openai/codex
 [Google ADK]: https://github.com/google/adk-python
 [Hermes]: https://github.com/NousResearch/hermes-agent
+[MiMo-Code]: https://github.com/XiaomiMiMo/MiMo-Code
+[Magic Context]: https://github.com/cortexkit/magic-context
+[Aider]: https://github.com/Aider-AI/aider
+[Roo Code]: https://github.com/RooCodeInc/Roo-Code
+[Continue]: https://github.com/continuedev/continue
+[OpenHands]: https://github.com/All-Hands-AI/OpenHands
+[Plandex]: https://github.com/plandex-ai/plandex
 
 ## License
 

package/dist/index.js

@@ -432,9 +432,6 @@ var PluginState = class {
     const last = this._lastMemoryNudgeMessageCount.get(sessionID);
     return last != null ? currentMessageCount - last : Number.POSITIVE_INFINITY;
   }
-  setModelContextWindow(tokens) {
-    if (tokens > 0) this._modelContextWindow = tokens;
-  }
   getModelContextWindow() {
     return this._modelContextWindow;
   }
@@ -15369,14 +15366,13 @@ function extractInputTokensFromMessages(messages) {
     for (const part of msg.parts) {
       if (typeof part !== "object" || part === null) continue;
       const p = part;
-      if (p["type"] === "step-finish") {
-        const tokens = p;
-        const input = tokens.tokens?.input ?? 0;
-        const cached2 = tokens.tokens?.cached ?? 0;
-        const total = input + cached2;
-        if (total > best) best = total;
-        if (best > 0) return best;
-      }
+      if (p["type"] !== "step-finish") continue;
+      const tokens = p;
+      const input = tokens.tokens?.input ?? 0;
+      const cached2 = tokens.tokens?.cache?.read ?? 0;
+      const total = input + cached2;
+      if (total > best) best = total;
+      if (best > 0) return best;
     }
   }
   return best;
@@ -16182,29 +16178,6 @@ function createNotifyHandler(client, logger) {
   };
 }
 
-// src/shared/model-limits.ts
-var KNOWN_MODEL_LIMITS = {
-  "deepseek-v4-pro": 1e6,
-  "deepseek-v4": 1e6,
-  "deepseek-v3": 64e3,
-  "deepseek-r1": 64e3,
-  "claude-opus-4": 2e5,
-  "claude-sonnet-4": 2e5,
-  "gpt-4o": 128e3,
-  "o1": 2e5,
-  "o3-mini": 2e5,
-  "gemini-2.5-pro": 1e6,
-  "gemini-2.5-flash": 1e6,
-  "qwen-max": 131072
-};
-function lookupModelLimit(modelID) {
-  if (KNOWN_MODEL_LIMITS[modelID]) return KNOWN_MODEL_LIMITS[modelID];
-  for (const [key, limit] of Object.entries(KNOWN_MODEL_LIMITS)) {
-    if (modelID.includes(key)) return limit;
-  }
-  return void 0;
-}
-
 // src/extract/enrich.ts
 import { stat } from "fs/promises";
 
@@ -16536,14 +16509,10 @@ var deepMemoryPlugin = async (input) => {
       const defaultModel = configResult.data?.model;
       if (typeof defaultModel === "string" && defaultModel.includes("/")) {
         const slashIdx = defaultModel.indexOf("/");
-        const providerID = defaultModel.slice(0, slashIdx);
-        const modelID = defaultModel.slice(slashIdx + 1);
-        state.recordFallbackModel({ providerID, modelID });
-        const limit = lookupModelLimit(modelID);
-        if (limit) {
-          state.setModelContextWindow(limit);
-          logger.debug("resolved model context window", { modelID, limit });
-        }
+        state.recordFallbackModel({
+          providerID: defaultModel.slice(0, slashIdx),
+          modelID: defaultModel.slice(slashIdx + 1)
+        });
       }
     }).catch((err) => {
       logger.debug("config.get failed, dream/distill will omit model", {