How Inkfold works, in everyday examples.

This page walks through what Inkfold does in four everyday situations: planning a trip, a long research or writing project, a freelancer working with several clients, and a small business.

Planning a trip

A trip to Japan in April: dates, budget, dietary needs and travel preferences settled once, available to whichever model answers next.

A long research project

Sources, notes and half-finished arguments carried from one session to the next, months apart, without repeating the background every time.

A freelancer with several clients

Each client's tone, constraints and history kept separate, so a question about Client A never leaks context from Client B.

A small business

Pricing, policies and standard answers settled once and reused by everyone on the team who asks a model for help.

Video · 104 seconds · captions

Inkfold in 104 seconds

How Inkfold works: the shared memory, the storage modes and where Inkfold stands today. The presenter is generated by a model, not a real person.

What Inkfold remembers, and how you correct it.

Inkfold keeps what matters from your conversations: projects, preferences and the facts you have settled, each with its source. It links people, places and topics into a map of your context — the context map — so a question next month starts from what you already said instead of from nothing.

The trip to Japan: you tell Claude the budget is about $3,000 and everyone prefers trains to flights. Weeks later, Gemini drafts a day-by-day plan and already knows both. If a detail changes, one click marks the memory wrong, outdated, or only for this trip — the correction becomes the new record, not a deleted one.

The right memories reach each question automatically.

You do not pick which memories to include. Inkfold looks at the question, searches the context map for what is relevant, and adds only that — within a size budget, so the model gets a focused brief rather than your entire history.

The freelancer: ask about "the contract clause for Client A" and only Client A's facts are added — Client B's project never enters the context, because the question did not call for it.

Your own keys, or one key for every vendor.

Inkfold works with the major model vendors: OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI and more. Add your own key for each, or a single OpenRouter key that reaches all of them. A spending view shows what each model has cost you, with budget rules you set. Automatic model choice by task and cost is coming — today, you choose.

The small business: a spending view shows that everyday questions are answered cheaply, while the model asked to draft a client proposal costs more — and a budget rule can cap the month before it surprises anyone.

For questions that matter, more than one model answers.

Four modes put several models to work on the same question, each suited to a different kind of decision.

Compare

Answers side by side

Example: the researcher asks the same question of two models and reads both summaries before deciding which citations to trust.

Council

Several answers, gathered into one

Example: the small business asks for a refund policy and Council merges several models' drafts into a single, considered version.

Debate

Models challenge each other

Example: the freelancer asks whether a contract clause is standard, and Debate has models argue it out until they agree.

Pipeline

Plan, draft, check, polish

Example: the trip itinerary goes through a pipeline stage that plans the days, drafts the details, checks the plan for gaps, and polishes the wording.

A better question, before it is sent.

The composer can improve, rephrase or tune a prompt for the model you picked, so a rushed question becomes a clearer one. A shared library holds ready-made personas, playbooks and routines that people publish, rate and adapt — such as a research playbook that goes from sources, to citations, to a summary. The library is in beta.

See what is in the library →

Three modes, chosen per conversation, free on every plan.

The mode is chosen per conversation, and no mode is behind a paid plan.

Smart · default

Best recall

Inkfold stores and indexes conversation content on its servers, encrypted at rest with a key Inkfold manages, so it can find the right context for you.

Private

Content stays unreadable to us

Content is encrypted in your browser with a key derived from your passphrase before it is stored, so Inkfold cannot read it and memory features are limited.

Incognito

Nothing kept

Nothing from the conversation is stored after it ends.

A workspace and memory a small team shares.

A team can share a workspace and its memory, so a decision settled once by one person reaches the rest of the team's questions. Team workspaces are in beta.

The small business: the owner settles the refund policy once; every teammate's questions about refunds start from that same answer.