A researched piece
Paste in a pile of article links. Let systems read and summarize all of them. Draft against those sources, edit the draft in your own editor at home, then run a fact-check pass whose corrections you approve one by one.
Ergoki
A place to keep the context your work depends on, so you never have to assemble it twice.
You paste the same background into a new chat. Again. You hunt down the same six links. Again. Last week’s good answer is buried in a conversation you can’t find, and the model has no idea what you already decided.
The thinking is not the hard part. The keeping is.
Every new chat starts from nothing. Ergoki keeps your sources, lists and notes in one living document — so the next question starts with everything the AI already needed to know.
Everything you make lives in a single database file you can download whenever you like.
A small example
Say you keep track of films — what you loved, what you walked out of, what your partner refuses to sit through. Here that is a dictionary: a table you define yourself, one row per film, with whatever columns you care about.
You add a few dozen. Then you point a system at them and the AI fills the gaps — genre, year, runtime — across every row at once, not one chat at a time.
Now drop that list into a construct: a document you build out of parts. Add one line at the end asking for tonight’s pick. Copy the whole thing into whichever AI you already use, and the answer comes back knowing your taste, because it genuinely does.
Next Friday you change that last line. Everything above it is still there. That is the entire idea, and it scales from film night to a research report.
Construct — “Movie night”
1 · Text
What we like, what we avoid, how much time we have.
2 · Dictionary
Films — 214 rows: genre, year, runtime, verdict
3 · Prompt
“Under two hours, nothing we’ve seen, good for both of us.”
4 · Result
— paste the answer back here —
Copy the construct → paste into any AI → paste the reply back in. Next time, only the last part changes.
A new account comes with that example already set up — a small dictionary, a system to run on it, and a Getting Started construct to follow. Fifteen minutes tells you whether the idea fits how you think.
The four pieces
Four words to learn. Everything else in Ergoki is built out of them.
Dictionaries
Tables you define. Films, sources, clients, card designs, diary days — you choose the columns.
Entries
The rows. Start one with almost nothing — a URL, a half-remembered title — and fill it in later.
Systems
Do one thing to all the rows at once: read the page behind a link, summarize, classify, tidy up.
Constructs
The document everything lands in: notes, lists, prompts, drafts. It grows as the work does.
The same loop, on real work
A hundred images, one style
Card art, game assets, character portraits. Change a single style line — “90s comic book”, “16-bit pixel art” — and regenerate the whole set so every piece still matches every other piece.
Everything about you
Tastes, allergies, the kids’ ages, how your team writes, what you already rejected and why. Written once, attached to any question, for years.
The short version
Spreadsheets are very good at rows and columns. Chatbots are very good at one-shot answers. Almost nothing is good at the messy middle where most real work actually happens: a job that takes a week, comes back in rounds, needs twenty sources you gathered on Tuesday, and ends with a human deciding what is good enough.
Ergoki lives in that middle. You collect information as entries in dictionaries you design, process them in bulk with systems, and assemble the results into constructs — documents that keep growing instead of being thrown away when the chat window closes.
It is built and maintained by one person who uses it for their own work every day: research pieces, creative catalogues, game design notes, diaries, standing briefings for recommendations. That shapes what it is. There is no roadmap driven by investors, no feature that exists to raise a metric, and no plan to make the free version worse later. The flip side is equally plain: it is one person’s tool, opened up to anyone who works the same way, not an enterprise suite with a support desk.
Made by one person, used every day
No investors to answer to, no growth targets, no quarterly reason to make it worse. It is built by someone who has to live with it tomorrow morning.
You keep the file
All your content sits in one database file. Download it whenever you want, or upload one to replace everything. Leaving is a button, not a support ticket.
Whatever AI you already use
Systems can call AI directly, but the plainest path is copy out, paste back. Nothing here is welded to one model or one vendor.
Four opinions it is built on
- 1. Work produces documents — information gathered, changed, and arranged. The help you want at each of those steps is different, so the tools for them should sit in one place.
- 2. Collecting the material usually happens long before anyone says “now write it.” A tool that only shows up at the writing stage arrives too late to be much use.
- 3. The work worth doing still needs someone who knows what “good” means here. Better to be handed a draft you can check than a finished thing you have to trust.
- 4. Sorting a list and prompting a model are the same kind of act — you are processing data — so they belong on the same shelf, next to scraping, filtering and find/replace.
Building blocks
Dictionaries & entries
Dictionaries are schemas for a kind of information (labels = columns). Entries are the rows. Movies, news sources, card designs, NPC dialogue, DJ track lists, diary days — open-ended.
Systems
Reusable processors over entries: scrape a URL, call AI to fill labels, regenerate artwork prompts, extract only what matters. Run one entry or a whole dictionary in batches.
Constructs
Ordered documents of text, lists, attached dictionaries/filters, tools, and chat. Copy the whole construct into an external chatbot, paste results back into a field, change the prompt, and go another round. Constructs get richer as the project progresses.
Prompt library
Store the prompts you reuse: research passes, first-draft instructions, final review with JSON find/replace output. Use them inside Ergoki or copy them out.
Vault & local sync
Turn construct text into Markdown vault files and sync with your computer. Edit in Neovim (or anything else); the construct stays current. Dictionaries and prompts can sync as packages too.
AI characters
Personas with rules and shared constructs as their memory. Useful for specialized helpers. This area is still growing; the core product does not depend on it.
Example workflows
How the pieces fit in practice.
1. A news report, end to end
- Add article URLs as dictionary entries.
- Run systems: scrape full text → categorize/summarize → extract only the facts that matter.
- Create a construct from a report template (intro, prompt slot, notes, attachments).
- Attach the processed sources; drop labels you do not need; copy the construct into an AI chat.
- Paste research back into the construct; swap the prompt to “write a first draft”; copy again.
- Promote the draft to a vault file, edit locally, keep the construct in sync.
- Final review prompt returns find/replace JSON (fact-check, answer in-text
!!questions!!, fix typos); apply it inside Ergoki.
Seven steps, one place. Nothing gets re-gathered, nothing gets re-pasted, and at the end you can still point at any sentence and say where it came from.
2. Consistent creative batches
Card art, game assets, or dialogue lines live as dictionary entries with shared prompt fragments. Change one global style instruction and re-run systems across the set so every piece follows the same palette and pattern.
3. Permanent context packs
A construct of movies you love / like / dislike / plan to watch, plus family tastes and constraints, becomes a reusable briefing. The request is only the last line (“recommend a horror film for tonight”); the AI already has the rest.
Map of the app
Your Nexus hub: bookmarks, folders, wallpaper, daily reminders, and a command bar for jumping around.
Uploads, generated images, and exports. Dictionaries and constructs can become Markdown or packages.
A record of what you did: system runs, AI interactions, personal log — activity next to your own notes.
Saved queries over dictionaries (e.g. only horror movies) you can reuse and attach to constructs.
Leaving is a button
Everything you make sits in one database file that belongs to you. From Settings you can download it whenever you like, or upload a different one and replace your account wholesale. Dictionaries, constructs and Markdown all export on their own too. A tool should hold your work while you are using it, not hold it hostage when you stop.
Is this for you?
Worth being straight about, before you spend an evening on it.
Probably, if
- You already use AI most days and feel the friction of starting over
- Your work comes back in rounds — research, writing, catalogues, worldbuilding
- You would rather assemble your own tool than accept someone’s workflow
- You want to be able to walk away with everything you made
Probably not, if
- You want an agent that takes the whole job away unattended
- You need shared team workspaces, roles and admin controls
- You want one button per task rather than pieces you arrange yourself
- A one-off question to a chatbot already answers everything you need
Chipping in, if you want to
Running this costs money — servers, and every AI call anyone makes. If you would like to help cover that, you can send any amount of ETH to the address below.
It is a donation and nothing else: it buys no subscription, it is not account credit, and it cannot be refunded — transactions on Ethereum are final and no one, including me, can reverse them. Only send what you are happy to give away.
Ethereum address · mainnet only
0xeD154AF09bB21b00E2a16356ae17738657174114
Anything sent on another network, or any token that is not ETH, may be unrecoverable.
If you send it from the same wallet you use to sign in with MetaMask, I can tell which account it came from, and a thank-you gift will show up on your settings page. Donations from any other wallet are just as welcome — there is simply no way to connect them to you.
Fifteen minutes will tell you
A new account already has the film example set up — a dictionary with a few rows, a system to run over them, and a Getting Started construct that walks you through one full round.