Field Notes
Short essays and working notes on persistent memory, agentic tools, and the slow craft of an assistant that remembers. Written for people who'd rather read than scroll.
Featured
We rewrote Ditto's agent + memory harness in Rust, open-sourced it under AGPL-3.0 (with a commercial license for partners), and built a self-contained starter kit so anyone can run the agent, practice memory retrieval, and prepare to mine on DittoBench (Bittensor SN118). A heads-up: this harness isn't in Ditto's production path yet — we're migrating to it by end of month, alongside the on-chain mining launch.
Read72 notes
A benchmark for any agentic harness that wields tools and memory — not just SWE. DittoBench clocks 10 models on whether they call the right tool, find the right memory, and how fast they do both, with a composite speed score that puts latency on the same footing as quality.
ReadHow Ditto's Seed Memories v4 finds the right memory from a vague question: a per-user query adapter and a subject graph that together lift Recall@1 by 7.6 points, train on CPU in seconds, and stay under a megabyte per user.
ReadDitto Code lets anyone build, deploy, and manage full web applications from a chat message. No coding experience, no DevOps, no deployment pipelines. Just describe what you want and watch it come to life.
ReadNobody fully understands why we dream. One theory is memory consolidation - your brain replaying the day and weaving it into what you already know. We built the algorithmic version of that for Ditto. The bugs we shipped along the way? Those were the nightmares.
ReadWe rebuilt the model that decides which past memories Ditto shows the AI. It now considers seven signals instead of three, looks at seventeen hints about your situation instead of six, and was trained directly on the ranking task — roughly doubling how often it surfaces the right memory.
ReadData scientists juggle experiments, datasets, feature pipelines, and model iterations across weeks-long projects — but their AI forgets every hypothesis between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a research partner that knows your entire analytical context.
ReadProduct managers juggle roadmaps, user feedback, stakeholder alignment, and competitive intelligence across dozens of features — but their AI forgets every strategic decision between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a product partner that knows your entire product context.
ReadReal estate agents manage dozens of clients, hundreds of showings, and thousands of property details simultaneously — but their AI forgets every client preference between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a real estate partner that knows your entire book of business.
ReadRecruiters and HR teams juggle dozens of open roles, hundreds of candidates, and endless interview notes — but their AI forgets every candidate between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into an HR partner that knows your entire talent pipeline.
ReadLawyers juggle dozens of active matters — statutes, deadlines, client instructions, opposing counsel's arguments — but their AI forgets it all between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a legal assistant that knows your entire caseload.
ReadTherapists and counselors carry rich context about every client — breakthroughs, setbacks, recurring themes, treatment goals — but their AI forgets it all between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a clinical support tool that knows your caseload.
ReadStudents juggle five courses, dozens of deadlines, and hundreds of concepts per semester, but their AI forgets everything between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a study partner that actually knows your coursework.
ReadSales reps juggle dozens of active deals, each with unique stakeholders, objections, and timelines — but their AI forgets every prospect between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a sales partner that knows your pipeline as well as you do.
ReadDesigners juggle brand guidelines, client feedback, design systems, and multi-project workflows — but their AI forgets everything between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a design partner that knows your aesthetic, your clients, and your process.
ReadMarketers juggle campaigns, audiences, brand voices, and analytics across dozens of channels, but their AI forgets everything between sessions. Here's how persistent memory, threads, and a knowledge graph turn AI into a marketing partner that knows your brand as well as you do.
ReadSmall business owners juggle customers, vendors, inventory, marketing, and bookkeeping, but their AI forgets everything between sessions. Here's how persistent memory, threads, and a personal knowledge graph turn AI into an operations partner that knows your business as well as you do.
ReadParents juggle schedules, allergies, school events, and milestones for every kid, but their AI forgets it all between sessions. Here's how persistent memory, threads, and a personal knowledge graph turn AI into a family assistant that actually knows your family.
ReadTeachers manage hundreds of students, adapt lessons on the fly, and track progress across entire school years, but their AI forgets everything between sessions. Here's how persistent memory, threads, and a personal knowledge graph turn AI into a teaching partner that never loses context.
ReadProject managers juggle dozens of decisions, stakeholder conversations, and shifting timelines across multiple projects, and their AI forgets all of it between sessions. Here's how persistent memory, threads, and a personal knowledge graph turn AI into a project partner that never loses context.
ReadCoaches and consultants manage dozens of client relationships, each with unique goals, history, and progress, but AI forgets everything between sessions. Here's how persistent memory, dedicated threads, and a personal knowledge graph turn AI into a practice partner that never loses a client's thread.
ReadDevelopers juggle multiple codebases, architectural decisions, and debugging threads, but AI forgets everything between sessions. Here's how persistent memory, dedicated threads, and a personal knowledge graph turn AI into a pair programmer that actually remembers your stack.
ReadYou're tailoring resumes, prepping for interviews, and tracking dozens of applications, but your AI forgets everything between sessions. Here's how persistent memory, dedicated threads, and a personal knowledge graph turn AI into a career partner that knows your story.
ReadContent creators juggle brand voice across platforms, recycle ideas without realizing it, and lose brilliant angles in throwaway AI chats. Here's how persistent memory, dedicated threads, and a personal knowledge graph turn AI into a creative partner that actually knows your content.
ReadFreelancers juggle dozens of clients, each with their own context, preferences, and history, but AI forgets everything between sessions. Here's how persistent memory, dedicated threads, and a personal knowledge graph turn AI into a business partner that never drops context.
ReadResearchers juggle dozens of papers, competing frameworks, and months of evolving analysis, but AI forgets everything between sessions. Here's how persistent memory, knowledge graphs, and context-rich threads turn AI into a research partner that keeps up with your thinking.
ReadYou're juggling investor updates, product decisions, hiring, and a dozen Slack threads, and your AI forgets everything between sessions. Here's how persistent memory, threads, and a personal knowledge graph turn AI into a co-founder that actually keeps up.
ReadWriters use AI for brainstorming, worldbuilding, and drafting, but every session starts from zero. Here's how persistent memory, threads, and knowledge graphs turn AI into a creative partner that actually remembers your story.
ReadYou're using AI to learn new skills, but every session starts from zero. Here's how persistent memory, knowledge graphs, and context-rich threads turn AI into a study partner that actually keeps up with you.
ReadTraditional journals gather dust. AI journaling with Ditto captures your thoughts automatically from conversations, text or voice, and builds a searchable, interconnected record of your thinking over time.
ReadYou explained your project, your preferences, your entire tech stack, and the AI forgot all of it by the next session. Here's why AI amnesia happens and how persistent memory changes the game.
ReadMost AI assistants handle one prompt at a time. Ditto spawns specialized sub-agents, research, image generation, web search, that run concurrently inside your threads, with full transparency and user-controlled guardrails.
ReadMost AI agents start every task from scratch. Ditto's agents, web search, image generation, and multi-step workflows, tap into your persistent memory to deliver results that actually fit your context.
ReadDitto connects directly to Gmail, Google Calendar, Docs, and Sheets, so you can search emails, schedule meetings, create documents, and update spreadsheets through natural conversation. All with persistent memory.
ReadDitto now integrates directly with Google Workspace. Connect your Gmail, Google Calendar, Drive, Docs, Sheets, and even Google Home, all from a single settings screen. Your AI assistant just got access to your real life.
ReadDitto now supports live voice conversations with AI, powered by Gemini's realtime API, client-side voice activity detection, and persistent memory. Every voice conversation is remembered.
ReadA memory-first AI raises a fair question: who controls the memory? Ditto's answer: you do. Here's how we handle privacy, data ownership, and user control.
ReadDitto is now a native app on iOS and Android. Your persistent AI memory, knowledge graph, and voice conversations, all in your pocket with native performance.
ReadDitto's Live Mode lets you have real-time voice conversations with AI, hands-free, natural, and backed by persistent memory. Your AI finally remembers what you said out loud.
ReadYour best AI interactions are buried in endless chat history. Ditto's Bookmarks and Collections let you save, organize, and instantly retrieve the conversations that actually matter, like a personal library for your AI knowledge.
ReadEvery developer has re-explained their project setup to an AI assistant. Ditto remembers your stack, your architecture decisions, and your debugging history, so you never start from scratch.
ReadMost AI assistants forget your files the moment the conversation ends. Ditto stores images, PDFs, and voice input as persistent, searchable memories, so your AI always has the full picture.
ReadAI conversations degrade as they grow longer. Ditto v0.26.0 introduces pre-computed memory summaries that keep responses fast, accurate, and affordable, no matter how long the conversation.
ReadMost AI assistants are a black box, you never know what context shaped the response. Ditto shows you exactly which memories it retrieves for every answer, with expandable inline cards and pre-computed context summaries.
ReadMost AI assistants use your memories silently, you never know which ones influenced a response. Ditto shows you exactly which memories were retrieved, why they were chosen, and how they scored.
ReadGPT for writing, Claude for code, Gemini for research, but three subscriptions, three chat histories, and zero shared memory. Ditto lets you use every model in one assistant, with one persistent memory.
ReadYour AI conversations are full of decisions, insights, and breakthroughs, but they vanish when you close the tab. Ditto captures everything into a searchable, visual second brain that compounds over time.
ReadClaude is great for code. GPT writes better marketing copy. Gemini crushes research. So why are you using the same model for everything? Here's how to match models to tasks, and how Ditto lets you switch between them without losing context.
ReadMost AI assistants forget your goals the moment you close the tab. Ditto injects your active goals into every conversation, so your AI always knows what you're working toward.
ReadEvery AI has chat threads now. Only Ditto lets you attach subjects, memories, and notes to yours, creating persistent, context-rich conversations you can return to anytime.
ReadA practical guide to getting started with Ditto, the AI assistant that remembers everything. Set up your first thread, build your knowledge graph, and connect your other AI tools via MCP.
ReadA practical comparison of how Ditto, ChatGPT, Claude, and Gemini remember your conversations. We break down visibility, control, portability, and what each approach means for your workflow.
ReadMost AI assistants treat every user the same. Ditto analyzes your Big Five, MBTI, and DISC profiles from your conversations, then adapts its responses to match how you actually think.
ReadSee how Ditto automatically extracts subjects, people, and ideas from your conversations and connects them into an interactive knowledge graph, a visual map of everything you've discussed with AI.
ReadConnect Ditto to any MCP-compatible service: Zapier, GitHub, Notion, databases, and more. Your AI assistant just got access to the entire MCP ecosystem.
ReadLearn how to integrate your Ditto memories with Claude, Codex, Cursor, and any MCP-compatible AI assistant. Step-by-step guide with OAuth and API key setup.
ReadWe trained a lightweight MLP to dynamically weight semantic similarity, recency, and frequency signals for memory retrieval, achieving 98.8% intent accuracy with sub-millisecond inference.
ReadHow we structured our SolidJS app for optimal code splitting - Suspense boundaries, lazy components, and the patterns that actually work.
ReadHow we used Knip to find and remove dead code, unused exports, and orphaned files - and what we learned about code hygiene.
ReadHow we identified the queries blocking our initial render and shaved hundreds of milliseconds off load time.
ReadThe testing methodology behind our 7.6x faster script execution claim - including the gotchas that almost gave us wrong numbers.
ReadThe iOS and Android issues we hit building a PWA - safe areas, camera stability, viewport height bugs, and why each platform breaks in unique ways.
ReadHow we made our PWA work offline by persisting Solid-Query cache to IndexedDB - the setup, gotchas, and performance wins.
ReadHow we deleted 80% of our scroll code by studying ChatGPT's DOM structure and discovering the power of CSS overflow-anchor.
ReadBuilding loading states that feel fast and don't cause layout shift - our architecture for skeletons, Suspense, and progressive loading.
ReadA catalog of SolidJS mistakes we made after migrating from React - prop destructuring, early returns, effect timing, and more.
ReadThe patterns, anti-patterns, and surprises we discovered after living with SolidJS in production for three weeks.
ReadHow we migrated 100+ components from React to SolidJS, achieved 7.6x faster script execution, and automated the entire testing process with Claude Code.
ReadWe rebuilt Ditto from the ground up. The result? 50% less memory, 42% faster loading, and a phone that stays cool.
ReadThe untold story of how a team of engineers built personal AI that remembers you, five years, dozens of prototypes, and one patent application ago.
ReadMeet Ditto, the AI assistant designed to remember, learn, and grow with you. Discover how personalized AI can transform your daily workflows.
ReadDive deep into the technology behind Ditto's memory systems and learn how persistent AI memory can transform your productivity.
ReadMaximize your productivity with these proven strategies for working effectively with AI assistants like Ditto.
ReadNothing filed under this topic yet.
