Vectorize

"Just give your agent memory" is easy to say. Here's what it actually takes. The moment you call Hindsight's retain, one sentence gets pulled apart into facts, resolved entities, a knowledge graph, and consolidated observations. Check it out 👇

Devin Desktop Persistent Memory (Formerly Windsurf)! Devin Desktop is the editor Cognition rebranded from Windsurf (formerly Codeium) in June 2026. The name changed; the gap didn't. Devin reads your codebase and holds a plan within a session, but it carries nothing across sessions. Close the editor, reopen it tomorrow, and the agent is a fresh model again, with no memory of the decision you talked through last week or the convention you set on Tuesday. The hindsight-devin-desktop integration adds persistent long-term memory to Devin. It's worth understanding how it gets there, because Devin Desktop doesn't expose lifecycle hooks to third parties. There's no place to bolt a sessionStart recall or a stop retain. Instead the integration uses two things the editor does support: remote Model Context Protocol (MCP) servers and always-on workspace rules. Know more: linkedin.com/pulse/devin-de…

Hermes Agent on Windows: Set Up Persistent Memory with Hindsight. Earlier this week, Nous Research announced that Hermes Agent is now natively supported on Windows. Here's how to give it persistent memory. Hermes Agent is a self-improving assistant with 40+ tools — chat, research, voice, gateway integrations into Telegram, Discord, Slack, and more. Out of the box, every conversation starts from zero: Hermes doesn't remember what you discussed yesterday, what project you were planning, or what preferences you've already shared. Hindsight closes that gap. Once configured, Hermes recalls relevant facts before every reply and retains the conversation afterward — across sessions, across platforms, across restarts. This post is the platform-neutral Windows guide. If you specifically want a coding assistant with codebase memory, see the Hermes Coding Assistant on Windows companion. Know more: linkedin.com/pulse/building…

@paulgauthier's Aider forgets your project every session. Not anymore. Tell it once. It remembers for good. Drop-in wrapper: pip install hindsight-aider

@_saranshbarua @NousResearch Good luck! We endorse shooting for the moon.

I’m planning to use Hindsight with Hermes for a highly ambitious moonshot project at work. I’m hoping I made the right decision. @Vectorizeio @NousResearch, please don’t let me down, or I’ll be fired!

Voice Agents That Remember: Adding Memory to Vapi with Hindsight. Vapi makes it remarkably easy to ship a production phone agent — managed STT, streaming TTS, telephony, transfers, and an LLM loop, all behind a single dashboard and API. The part it leaves to you is the memory. Every Vapi call starts from zero. The caller who explained their account problem yesterday explains it again today. The customer who booked an appointment three weeks ago has to spell their name again. The agent that diagnosed a routing issue last Tuesday rediscovers it on Wednesday. Persistent memory across calls isn't a feature you can turn on inside Vapi — but it's a feature you can drop in with a webhook. The hindsight-vapi integration adds long-term memory to any Vapi assistant with one HTTP endpoint. Memories are recalled at the start of every call and the transcript is retained at the end — no per-turn integration, no changes to your assistant config, no custom tool wiring. Know more: linkedin.com/pulse/voice-ag…

Love that you can now run Hermes Agents in the @NousResearch Portal with just a few clicks. You can add Hindsight memory too. Here's how👇

Long-Term Memory for Google ADK Agents with Hindsight. Google's Agent Development Kit gives you a clean abstraction for building Gemini-powered agents — sessions, runners, tools, plugins, multi-agent composition. It has a memory interface too: BaseMemoryService, the contract any persistent-memory backend can implement. What it doesn't ship is a memory backend. That's left to you. hindsight-google-adk is the drop-in implementation. Wire it into your Runner, and sessions are automatically retained when they end. The next time the agent calls search_memory, results come back from a Hindsight bank scoped to that user. No retrieval code to write, no vector store to manage, no schema decisions to make up front. Know more: linkedin.com/pulse/long-ter…

Watch it remember 🧠 Teach Zed's agent a fact once, then open a brand-new thread. It recalls it, no reminder, no re-explaining your codebase every session. @zeddotdev @nathansobo

@S_BatMan We did enjoy your articles and are glad you love Hindsight

Hah it takes me to today to realise @Vectorizeio is behind hindsight, hope you guys enjoyed my 10 articles which are basically Hindsight love letters :D As a connoisseur of agentic memory systems, i can say with confidence Hindsight is a Leader.

@Validate_QA @vercel @vercel_dev @v0 @rauchg it's bounded, not stuffed. recall is a fixed budget of the most relevant memories, injected once before the turn.

@Vectorizeio @vercel @vercel_dev @v0 @rauchg how much context are you stuffing in before each turn tho

Your @vercel eve agent forgets you every new chat. We fixed that in. Recall before each turn, retain after. The model never calls a memory tool. Teach it once, new session already knows. cc @vercel_dev @v0 @rauchg

@nickventuri @vercel @vercel_dev @v0 @rauchg that was the exact itch. now it reintroduces itself to you. new session, it already knows your setup!

@Vectorizeio @vercel @vercel_dev @v0 @rauchg having to reintroduce myself to my own code every five minutes was exhausting 🙄

How oh-my-pi Built Persistent Codebase Memory on Hindsight! oh-my-pi (or omp) is a TypeScript + Rust coding agent for the terminal — with 10,600+ stars, 40+ providers, 32 tools — it aims to have the IDE wired in, with each tool meticulously crafted for the agent's performance. When it comes to memory, they trusted Hindsight due to it's exceptional long-term recall performance and the multi-faceted retrieval technique used underneath. It's not a thin integration either: it features a complete memory subsystem built on top of Hindsight's API — bank scoping, mental-model seeding, debounced retain queues, auto-recall on the first turn. This post is a tour of how they built it, with code pulled straight from the public repo. If you're considering Hindsight for a coding agent of your own, this is a useful reference for the shape of a deep integration. Know more: linkedin.com/pulse/how-oh-m…