Empirical Blog

Product updates, implementation write-ups, and lessons from building portable AI memory.

• Everything we shipped in October 2026-10-31

  Writing Voice, Empirical Tokens, workspaces that know what they hold, files on a memory, a Daydream that finds real talks for you, and Empirical in more places. The full list, in one page.

• You told your agent about the lead. Then you typed it into the CRM again. 2026-10-26

  A CRM for leads. A todo list with reminders. A tracker for the work. Built out of what you already said to an AI, so nobody fills in a form twice. Here is what lands where, why the classifier is allowed to answer "the record does not say", and the three places it stops.

• The decision was in memory. The contract was in Slack. 2026-10-22

  Empirical memories can now carry the files they are about. Twenty per memory, 25MB each, and deleting one is reversible for thirty days with the purge date printed on the receipt. Here is what it does, and the two places it deliberately stops.

• An AI agent that learns from you is great. Who owns what it learns? 2026-10-07

  Empirical's Daydream emailed me an article about Hermes, an open source agent that learns skills from how you use it. It made me ask where a year of AI context actually lives, and what happens to it when you switch models.

• Anthropic published four weeks of optimization research. We turned it into a memory an agent could build on. 2026-09-18

  Anthropic's own scientists hand-optimized 30-plus open-source biology models and published all of it. We loaded that work into Empirical, pointed an agent at a model they'd never touched, and in under two hours it reused their work, and then found one thing their study never had a reason to look for.

• Memory that comes back to you: reminders in Empirical 2026-09-17

  Memory only answers when you ask it something, which is fine until a note has a date on it. Any Empirical memory can now carry a due date. When it comes up you get a calendar alarm, an email, and a heads-up inside whatever AI agent you happen to be working in.

• Blueprints now interview you before they install anything 2026-09-03

  A canned memory graph made your assistant sound like a template. The new guided setups interview you, branch on what you say, and turn the transcript into real memories your assistant uses from the first question you ask.

• Multi-agent workflows have an amnesia problem 2026-08-25

  Every agent in a pipeline starts from zero unless something manually threads context between them. Salesforce measured what that actually costs, and it is not small.

• The Empirical CLI is becoming the front door 2026-07-30

  Installing Empirical should feel like choosing where you want memory to work, not like learning the internals of an integration stack.

• Empirical became a set of tools, not just one plugin 2026-07-30

  The memory layer is getting smaller, clearer, and more useful: one core skill, plus optional tools for work history, hard-earned lessons, and project follow-ups.

• A Todo list that understands the project behind it 2026-07-30

  The new optional Todo skill keeps follow-ups attached to real projects, whether you are shipping software, finishing a book, or trying to make a personal goal concrete.

• Memory trust and portability are live — error codes, receipts, duplicates, and export 2026-07-28

  A reader's email about ChatGPT's memory quietly failing on him turned into a punch list. Here's what we shipped in response, and how to run it from ChatGPT, Claude, or a terminal.

• Your AI needs your second brain 2026-06-27

  If you already spent years capturing ideas, decisions, research, and notes, you should not have to re-explain all of it every time you open a new AI chat.

• Your vibe-coded app got serious faster than expected 2026-06-25

  The dangerous part was not that AI wrote code quickly. It was how easy it became to forget which shortcuts were only safe while the app was still a toy.

• Your AI coding agent needs scar tissue. 2026-06-13

  Context tells an agent what to do. Scar tissue reminds it what not to repeat.

• Have you seen this ChatGPT warning before? Here’s a fix. 2026-05-31

  I'd been keeping months of health notes in one ChatGPT thread. When I noticed the warning about long conversations, I started wondering what would happen to all that history over time. That question eventually led me to move the record somewhere more reliable.

• Empirical saved my ass. 2026-05-27

  Two weeks of work, gone in a single refresh. I knew it had to be somewhere. The AI didn't believe me until memory backed me up, and pointed right at it.

• Audit every memory recall before it shapes output 2026-05-20

  Inspecting what was recalled, why it ranked, and where it was injected makes memory behavior explainable and debuggable.

• Build on the Empirical memory API in minutes 2026-05-20

  A developer-grade memory API means MCP, REST, and TypeScript paths that are production-usable without custom glue.

• Portable memory with no vendor lock-in 2026-05-20

  Exportable memory gives you model freedom. Your context remains yours even when providers, tools, and workflows change.

• Keep work and personal memory separate with scoped memory 2026-05-20

  Scoped memory lets you isolate project, team, and personal context so each AI call sees only the slice it should.

• Time-aware memory relevance for changing work 2026-05-20

  Time-aware retrieval favors context that is both relevant and recent so your AI responses reflect current priorities, not stale history.

• Universal recall across every tool you use 2026-05-20

  Why one memory graph across ChatGPT, Claude, Cursor, terminals, and workflows removes repeated setup and keeps outputs consistent.

• Your coding agent is burning context you will never get back 2026-05-11

  Static instruction files load into every turn whether they are relevant or not. Switching to on-demand memory retrieval gives the same 5-hour window 22-45% more range.

• How I Used Codex + Empirical to Lock In My Writing Voice 2026-04-30

  I used Empirical with Codex to stop tone drift and define a repeatable voice path through guided questions and live rewrites.

• Workspace implementation is live: cleaner team memory, fewer context collisions 2026-04-29

  Workspaces are now first-class in Empirical, so team context stays separated without slowing down day-to-day recall.

• Discord bot implementation lessons: reliability first, features second 2026-04-26

  Real Discord usage exposed friction in joins and notifications, and those fixes directly shaped the workspace model.

• CLI install updates: Q&A onboarding with dynamic global prompt building 2026-04-20

  CLI setup now asks practical questions and generates a stronger default prompt policy without a long manual setup.

• Expanding full CLI tools alongside MCP integrations 2026-04-18

  Expanded first-class CLI tooling so teams can run memory workflows directly in scripts and terminals, not only via MCP clients.

• 100 users reached: signal and what comes next 2026-04-13

  Crossing 100 users provided clearer signal on what matters most: reliability, portability, and context ownership.

• Installing Empirical in Claude web app: docs that remove setup guesswork 2026-04-05

  Published a practical Claude web app install guide to reduce connector setup errors and improve first-run success.

• Submitting Empirical to the ChatGPT app ecosystem 2026-04-04

  Submitted Empirical for ChatGPT app approval and tightened tooling metadata, test cases, and reviewer reliability.

• HIPAA-aligned storage and auditing improvements 2026-03-20

  Shipped stronger storage and auditing controls for high-trust memory workflows that require tighter compliance posture.

• Blueprint concept: portable skills across AI tools 2026-03-15

  Blueprints package repeatable skills so users can move workflows across chat apps, coding CLIs, and agents without restarting.

• Geocoding and image processing: richer memory with location context 2026-03-15

  Added geocoding and image processing support to make memory entries more useful for location-aware recall.

• Why Empirical exists: context that survives tool switching 2026-02-02

  Empirical started from one frustration: valuable AI conversations were hard to carry from chat into execution tools.