In development · Early access 2026
A workspace where your past ideas resurface, your thinking develops, and everything you create can build on what came before.
Most teams that leave never reach a first clear result. In your March notes, 7 of 9 teams that left stopped before they finished setup.
The value appears too late. Show one result in the first session.
From chat "Why do teams leave?" · edited now
Goal: every new team sees one clear result in the first session.
Plus PDFs, e-books, web pages, highlights, voice memos and handwritten notes
Why Scrivel
They are spread across many tools and many models. A tool can shut down, change its terms or close your account. Then that work is gone.
Scrivel keeps a copy of all of it, in one place that you control.
By the numbers
25
Planned sources, from chat exports to handwritten notes
8
Agents, each with one clear job
4+
Model providers, plus any model you run on your own machine
100%
Of your work, exportable as plain files
01 · Own your work
Scrivel keeps its own copy of everything you make with AI. You choose where that copy lives, and which model reads it. You can take all of it with you at any time.
Scrivel keeps a copy of every item. Each item records its source and its date.
| Source | Type | Items | Status |
|---|---|---|---|
| CClaude history | Chat export | 1,284 | Imported Mar 2 |
| GChatGPT history | Chat export | 912 | Imported Feb 28 |
| GGemini history | Chat export | 140 | Importing 64% |
| DGoogle Drive | Cloud storage | 87 | Synced 1 h ago |
| NNotion | Work tools | 312 | Synced 1 h ago |
| SSlack | Work tools | — | Connect |
The work you make with AI is spread across many tools and models. If you lose a tool, you can lose that work. Scrivel calls this work your Generated Intellectual Property (GIP) and keeps its own copy of it.
You choose where Scrivel stores your work. You can export all of it as plain files at any time. Scrivel never uses it to train a model.
Choose the model for each task. Add your own API key from Anthropic, OpenAI, Google or OpenRouter, and pay the model company directly, at their price. Or run a local model on your own machine through Ollama, LM Studio or any OpenAI-compatible server.
Your knowledge base does not depend on one AI company.
Bring in the work you already have. Import your past AI chats from the export file that each chat app gives you. Connect cloud storage and work tools one time, and they stay in sync. Scrivel keeps each chat that you have in Scrivel, with any model.
Every item keeps a record of its source and its date.
Scrivel turns audio, video, images and handwriting into text, so you can search all of it.
Scrivel also works the other way. Through MCP, apps such as Claude and ChatGPT can search your Scrivel library, and add to it when you allow it.
Each release adds more sources.
02 · Find it again
Search by the words in an idea or by what you remember about it. Then look at the results in the way that helps you most.
Search and open your chats, ideas, notes, artifacts and projects in milliseconds. Search runs on your device, so results appear while you type.
The whole app responds at this speed, not only search.
You often remember the shape of an idea, not its title. Describe what you remember: the month, the length of the chat, the result, or that you did not finish it.
Scrivel finds the match. You do not lose an idea because you forgot its name.
Look at the same knowledge in different ways. A frame changes how results look: questions and answers, a timeline, or a list of decisions. A lens changes what you look for: only decisions, only open questions, or a skeptic's view of your notes.
Use them together to see things that a plain list of results hides.
While you work on something, Scrivel shows a few past ideas that relate to it. They appear at the moment you can use them. Add one to your current work in one step.
03 · Work in one place
Move between a chat and a document without a change of app. Arrange your panes the way you want. Save the path you took, and return to it later.
Chat and write in the same place, and move between them freely. Turn part of a chat into a document, or discuss a document in a chat. Ask the AI for help inline while you write documents, notes and ideas.
Scrivel saves each thing the AI makes as its own item. That item links to the chat that made it, so you can search for it and it can resurface later.
Sometimes you want to wander. You jump between ideas, open new chats, save quotes and phrases, ask follow-up questions and take notes. A trace records that whole path and saves it.
Return to a trace later. See how one idea led to the next, and continue from any step.
Put a chat, a document and its sources side by side. Panes open, close, expand and collapse. You arrange the workspace with almost no thought.
Make the knowledge base feel like yours. Set your own keybindings, change the colors, add images, and build dashboards that show what matters to you.
04 · Think for yourself
Some effort is worth it, because understanding and original ideas come from it. Scrivel keeps that effort in the places that matter to you.
Generative Friction is on. Write your own answer. Then Scrivel shows the AI's answer.
Hint: what happens to a memory each time you recall it with effort?
Your answer is correct. Recall with effort strengthens a memory more than reading it again. Spaced reviews also tell the brain that the fact is still important.
You missed one point: the gaps between reviews must grow.
Generative Friction is a set of switches. Each switch asks you to do part of the work yourself before the AI helps.
You choose which switches are on.
05 · Learn from your work
The Learning Studio makes practice from your own chats, notes and highlights. You recall, explain and argue, so the ideas stay with you.
Study setOnboarding research · 5 due
Why do most teams leave in the second month?
Your notes say
You were guessing and got 2 of 3. You know more than you think.You were fairly sure and got 2 of 3. That is good calibration.You were certain and got 2 of 3. Look again at the point you missed.
Recall first. Then rate how well you remembered.
Your claimLong setup is the main reason teams leave.
Your notes from Feb 2 say teams with a fast setup also left in month two. If setup is the main reason, why did they leave?
Seat or usage pricing?
For seats
For usage
Make the side you disagree with as strong as your own.
SocratesYou say people forget because they do not review. Do they forget everything at the same speed?
YouNo. Some things stay for years.
SocratesThen what is different about the things that stay?
You
Write both voices, or let Scrivel ask the questions.
Walk the rooms in your mind. Then recall the three reasons.
Each review comes just before you would forget.
Practice and test what you want to understand and remember. Scrivel makes the practice from your own chats, notes and highlights. The tools come from research in cognitive science and teaching. They keep you active while you learn.
06 · Agents
Agents do the upkeep and the deep research that people seldom do by hand. For harder problems, you direct several agents at once.
Agents work on your knowledge base where manual upkeep seldom happens. Each agent has a clear job.
Merge duplicates, repair links and keep things in order.
Between things you wrote or concluded.
Flag ideas that are out of date.
Distill, synthesize and combine related material.
Research and generate ideas in loops, toward a clear goal that you set.
Show where a small amount of work has a large effect.
Look at how you use Scrivel and suggest improvements.
Show how your ideas connect and how they changed over time.
07 · Compose a team
Scrivel reads your goal and recommends a team of agents for it. It chooses each role, its skills and its model, and shows you why. Change anything, then start the run.
Why do teams leave in month two? Find the main reasons, check each one against our interviews, then suggest two fixes.
Should we price by seat or by usage? Argue both sides with our data, then recommend one.
Tidy my notes on pricing: merge duplicates and repair broken links.
Recommended teamPlan → research in parallel → check → write
Why this team: Your goal has four parts and two kinds of source. Each one gets its own agent.
Other teams Scrivel considered
One agent1 agent · about $0.30
It does not check the web’s claims against your interviews.
Seven agents, with a review7 agents · about $3.10
Over your $2.00 budget. The goal asks for reasons, not a second review.
Why
Your goal says check each one against our interviews
Reports can be wrong about your users. This agent tests each claim against what your users said.
Skills
Model
Config · read-onlyThe orchestrator starts the run from this.
About $1.40 · about 9 min · 5 agents
Leaves this device: the goal, 12 interviews, the findings.
Nothing runs until you start it.
Run #46 started. 5 agents are at work.
See the runThis team costs about $3.10. Your budget is $2.00.
Describe a goal in your own words. Scrivel recommends the smallest team that does the job. A team of agents uses many times more tokens than one chat, so more agents is not better.
Each agent points to the words in your goal that caused it. Change a role, a skill or a model, or use a team that Scrivel considered. When you start the run, this team becomes its config.
One agent for each part of the goal.
A pipeline, parallel work, a check, a debate, or one agent.
From your skill library, or new for this goal.
A local model for private data. A strong model for careful checks.
What each agent can read, and what it cannot.
A cost cap for each agent and for the run.
When each agent is done.
An agent joins when two disagree. Scrivel asks you first.
08 · Many agents, one goal
Give a hard question to a team of agents. Each has one job and its own model. They share what they find, test each other's claims, and ask you when they disagree.
Why do teams leave in month two? Find the main reasons, check each one against our interviews, then suggest two fixes.
Leaves this device: the goal, 12 interviews, 5 findings. The rest stays here.
Needs youSettled by youTwo agents disagreeTwo agents disagreed
Web researcher Public reports say price is the main reason teams leave in month two.
Skeptic Your interviews name price 2 times in 12. Setup and ownership come up 9 times.
You kept both. The answer will note the gap.
SkepticClaude 4.5 SonnetLog
For harder problems, run several agents together and stay in control. Guide each agent, stop it, or give it a new direction while it works. Nothing changes your notes until you approve it.
Use several agents for deep research, for debates between agents that argue different positions, for parallel work, and for analysis from many sides.
09 · The agent harness
The agent harness holds all your agents in one view. The harness graph shows each agent, what it reads, and what it sends to the others. You steer them, and each choice has a reason that you can read. Replay the run from the start to see how it went.
$0.53
of $2.00
41k tokens · 10 min
Your notesOn this device
The webPublic reports
Interviews12 · leave this device
Draft answer3 changes wait for you
Activity26 events
ShowingAll agents
10:00 / 10:00
Scrivel harnesses many agents at once. See the run as a graph. Each agent is a node. The graph shows what each agent reads, what it sends, and to whom.
Each choice that an agent makes has its reason, the options it did not take, and its sources. Replay the run to see each step again, in order.
The basics
Replies appear while the model writes them. Stop a reply, edit and retry a message, attach files and copy text.
Documents and artifacts open in a pane next to the chat. You edit them live.
Your library and search work without an internet connection. With a local model, chat works offline too.
Scrivel saves your work automatically. You can undo any change. A log shows each change that an agent made.
Scrivel shows which content goes to which model.
Export everything as plain files at any time. Your work never depends on Scrivel.
Questions
Scrivel is a knowledge base for the work you do with AI. It collects your AI chats, notes, documents and artifacts in one place. You can search them, arrange them and use them again.
Claude, ChatGPT and Gemini each let you download your chat history as an export file. Scrivel will import that file. Scrivel will also sync with cloud storage and with work tools such as Notion, Slack and GitHub. You can add web pages, PDFs, e-books, highlights and newsletters, and also videos, podcasts, voice memos, photos, handwritten notes and files. Scrivel turns audio, video and images into text, so you can search them. Scrivel keeps each chat that you have in Scrivel, with any model.
No. Scrivel never uses your work to train a model. You choose where Scrivel stores your work. You can export all of it as plain files at any time.
Yes. Scrivel is model-agnostic. Use a frontier model from Anthropic, OpenAI or Google. Or use a local model through Ollama, LM Studio or any OpenAI-compatible server.
Not for a local model. A local model runs on your own machine and needs no account. For a frontier model, add your own API key from Anthropic, OpenAI, Google or OpenRouter. You pay that company directly, at their price.
Chat history stays inside one tool. Scrivel collects your work from all of your tools in one place. It finds ideas by what you remember about them, and it shows past ideas when they are relevant. You can export everything.
Scrivel is in development. Join the waitlist to get one email when early access opens. Scrivel sends no other email.