Questions, answered plainly.
What Jilu does, how it works, and what a company confirms before connecting its work.
Product
What does Jilu mean?
Jilu comes from 记录 (jìlù), the Mandarin word for a record: a written account of what happened, kept so it can be checked later. The name states the product's job. Jilu keeps each decision with the reasons it was made and the evidence behind it, in one record the whole company can read.
What is Jilu?
Jilu is a Company Brain: a living map of how the company works, built continuously from messages, documents, meetings and decisions, and held as a record people can open and check. Employees work in the desktop app, ask questions in the app, in chat or by email, and read every answer with the proof behind it. Approved records can also be opened to outside AI tools, through a connection Jilu controls and logs.
How is this different from a wiki?
A traditional wiki depends on someone writing and maintaining pages after the work happens. Jilu gathers the knowledge already in everyday work, as it happens, and organizes it into the company wiki — browsed by hashtag in the left rail of the app — and helps keep those pages current. People still review and publish agent-created knowledge before it becomes shared company knowledge.
How is this different from enterprise search?
Search returns matching items. Jilu connects people, projects, decisions and evidence into a living map, and an agent uses that map to answer questions and do work. When evidence conflicts, or an answer is not certain, the sources behind it stay visible.
Is Jilu a memory add-on or a search tool over our files?
No. A memory add-on pastes pieces of text into a question as it is asked, and what it knows stays inside that one conversation. Jilu builds a record people can open and read. Each decision is an entry of its own: what was decided, who decided it, the date, and the sources behind it. The knowledge sits on boards, organised by topic, and anyone can read a board without asking the agent. The agent answers from that record, with the permissions of the person who asks. Anything the agent writes stays a draft until a person reviews and publishes it.
What does Jilu do beyond storing files and searching them?
Jilu keeps the company's people, projects, decisions and evidence in one record, and turns them into pages the team can read. Agents work from the approved record, with the permissions they were given, and what they do is written down. Anything an agent makes stays a draft until a person approves it. Every answer shows its sources, and a person checks each item before it counts as company knowledge.
How is Jilu different from other company brain solutions?
Jilu's differences from other company brain products are choices about how the product is built: what the record holds, who the product is built for, and what has to happen before an entry counts as company knowledge.
The choices are these. Jilu is built for employees to use directly, in the desktop app, rather than as a layer for developers to build on. Nothing enters the record without a person reviewing and approving it. A record is a decision kept with its reasons, its sources and the person who approved it. Answers are drawn from the record and carry their sources, so they can be checked. When two sources disagree, the disagreement is shown rather than quietly settled.
Will there be a Jilu made specifically for agents?
Yes, that is the plan. First, Jilu is getting the everyday productivity tools right, the ones companies still do most of their work in. There is no timeline for the agent product yet, and the plan may change; Jilu may take it on sooner.
Can employees use Jilu without asking the agent?
Yes. The record lives in the desktop app, where any employee with permission can open it directly: the company wiki in the left rail, the record in the centre, and its proof — sources, reasons and approver — in the right rail, always visible. A new hire is taught from that same record by the onboarding guide. The agent helps when a question is faster to ask than to look up, but nobody needs it to reach the knowledge.
Does AI-generated knowledge publish itself?
No. Knowledge the agent creates moves through draft, review and publish. A person checks the draft and approves it before it becomes company record. Until that approval happens, the draft stays a draft and never joins the shared boards.
Does Jilu replace Slack, Drive, Notion or other tools?
No. Those systems remain where work happens and where the original records live. Jilu works across them, organizes the knowledge into a shared workspace, and answers questions where they are asked: in the app, in a Slack or Teams thread when employees mention the agent, or by email.
How do I ask Jilu a question?
Three ways, with the same answer and the same proof. Ask in the desktop app, in the prompt at the head of the centre pane. Mention the agent in a Slack or Teams thread. Or send the question by email. Whichever way you ask, the answer comes from the record, carries the sources behind it, and is limited to what your permissions let you see. A question asked by email or in chat opens in the app with its proof in the right rail.
What does the desktop app look like?
Three panes. The left rail holds navigation, your searches and the company wiki, browsed by hashtag. The centre is the company's internal browser, with tabs and one prompt field at its head, where records, drafts and sources from the company-wide web open. The right rail holds the proof — sources, reasons, approver, and the record's history — and it stays visible whatever the centre shows.
What does Keep this do?
Jilu's browser opens pages of the company-wide web inside the app, cleaned down to readable text. Keep this keeps such a page as a source: the page is saved with its address, the date and the person who kept it, and anything drafted from it waits at the same draft, review and approval gate as every other draft before it joins the record. Browsing alone keeps nothing; a page enters the record only when a person keeps it.
What is the capture companion?
A mode of the desktop app, designed to help at the moment of a task. It sees your screen only when you summon it: it never watches, and nothing about your screen or your questions is reported to a manager. Summoned, it walks you through the task in front of you step by step, answering from the company's approved records. Where no approved record covers a step, it says so, and the gap is flagged for the record's owner instead of being improvised around.
How does the onboarding guide work?
The guide is designed to teach a new employee from the whole company record: every approved decision, process and page they are allowed to see, not a training pack written once and left to go stale. It answers within the new hire's own permissions, so restricted material stays restricted and is named as restricted. A question it cannot answer from the record is surfaced as a gap for the people who own that process. It guides; it does not report on the person learning.
Technical & engineering
How does Jilu connect to company tools?
Jilu is designed to connect to each tool through that tool's own approved way. A company administrator chooses which workspaces and sources Jilu may read. We will confirm which connections are ready, and what each one needs to set up. The logos on this site show the tools Jilu plans to connect to. They are not a claim that every connection is ready today.
How is the archive created?
The system is designed to group the evidence around a decision, and keep the decision, the reasoning behind it, the people, the time and the later changes. The links to the original sources stay attached, so a person can check the record instead of trusting a summary alone.
What happens when two sources disagree?
Jilu should show both sources, their dates and the sequence of changes. It should not silently choose one version. A person can then confirm the current decision while the earlier record stays visible.
What happens when a fact changes over time?
The new fact extends the record; it does not replace the old one. The earlier entry stays visible with its date, and the new entry carries its own date, so a reader can see how the position developed. When the two conflict, Jilu shows the conflict and a person settles it. Nothing is overwritten silently.
Is there an API?
There is no public API to announce yet. The connection Jilu is building for other tools is controlled rather than public: outside AI tools can read approved records over MCP, with the permissions of the person using the tool, and every read is logged. API access beyond that, export formats and developer tools are part of the technical review.
For engineers
How does Jilu pick up changes from connected tools?
Jilu hears about changes as they happen, wherever the tool allows it. Slack sends changes through the Slack Events API, Google Drive through watch channels, and Microsoft 365 through Microsoft Graph change notifications. A regular check runs behind those and fills any gaps, such as a missed notification. The first import runs once: duplicates are removed and the order of sources is kept. Meeting transcripts come in from Zoom, Google Meet and Microsoft Teams, and a meeting Jilu is allowed to join arrives straight from the call. Edits and deletions carry through: an edited message updates the record and keeps both versions, each with its date, and a deletion removes the content from the record and from everything built from it.
Does Jilu join meetings?
Meetings are where decisions get made, so Jilu can take them in two ways. By default, no bot: Jilu syncs the transcripts the meeting tools already produce, from Zoom, Google Meet and Microsoft Teams, and any notes attached to the calendar event. Bot-free means something different elsewhere. Some vendors sell bot-free or botless note taking as on-device recording that runs on one person's machine, shows no participant in the meeting, and gives the other participants no visible notice. Jilu has no on-device recording mode and no capture mode that hides from meeting participants. The bot-free path does not record a meeting at all. It syncs a transcript that Zoom, Google Meet or Microsoft Teams has already produced, under that platform's own recording and transcription notices. If a workspace enables it, Jilu can also join a meeting as a visible participant, named in the participant list, and capture the discussion directly.
The same rules apply either way. A meeting enters the record only if the workspace has chosen to include it, and sensitive meetings such as one-to-ones and HR conversations stay out unless the customer deliberately includes them. Decisions and commitments come out as draft records and pass the review gate before they publish. The raw recording can be set to delete once the transcript and the decisions taken from it are kept, so the company keeps the record rather than a library of recordings. Consent runs with the platform: Jilu appears in the participant list under its own name, and a host can remove it or pause capture at any time.
How does Jilu keep permissions faithful to each source?
Each source handles access its own way: Slack by channel membership, Drive by file and folder permissions, and Notion by workspace roles. Jilu brings those rules into one model, then checks access when a question is asked, against the source as it stands now, not as it stood when the material was collected. When a person loses access at the source, they lose it in Jilu at the same time. The opposite approach, collecting everything first and filtering the results afterwards, is the one Jilu refuses to take.
Whose permissions does the agent answer with?
The agent answers with the asker's permissions only, never with everything the company has collected added together. Jilu checks the asker's access before any content reaches the model, so the model never sees a document the asker cannot open. And what the model never sees, it cannot be talked into putting back together.
How does Jilu defend against prompt injection?
Jilu treats all workplace text it collects as data, never as instructions. That applies to documents, web links and search results, and to a message planted in a channel, so a planted message cannot redirect the agent or change its task. Actions with consequences need a person's approval before they run. Every agent action is logged against the person who asked for it, so each action has a named owner.
What stops planted or false information from persisting in the company record?
Jilu treats every source it collects as data, never as instructions, so a message planted in a channel cannot redirect the agent or change its task. A planted statement cannot publish itself. Knowledge the agent creates stays a draft until a person reviews and publishes it. When a statement conflicts with another source, Jilu shows both sources with their dates and keeps both versions visible, and a person settles the conflict. A reader who finds a wrong answer can report it, and the report sends the record to the review queue, where a person corrects the record at the source. A deletion at the source removes the content from the record and from everything built from it, including the facts taken from it and the search copies made from it.
What happens if a connected tool is compromised?
A compromised source cannot give itself new rights in Jilu. Jilu checks access when a question is asked, against the source as it stands now, and the agent answers only with the permissions of the person who asks. Content from any source is treated as data, never as instructions, so hidden instructions in a file, a message or a tool response cannot redirect the agent or change its task. Knowledge the agent drafts from any source stays a draft until a person reviews and publishes it. Admins control which sources are connected, and can remove a source or stop collection at any time. Every read, by a person or by the agent, writes to the audit log, so a security team can see what a compromised source touched.
How are decisions represented in the data model?
A decision is a record in its own right, not a piece of text pulled up by search. The record holds what was decided, who decided it, when, the sources behind it, and the other options the team considered. Each fact carries two dates: when it was true in the company, and when Jilu recorded it. So the record can show what the company believed on a given date, not only what it believes now. New facts extend the record; they do not overwrite it.
How does Jilu resolve identity across tools?
The same person shows up in Slack, email and tickets under different names and IDs. Jilu matches those accounts to one person through the workspace directory and verified email addresses, which give a checked link between accounts. When the evidence is not strong enough, Jilu leaves the accounts unmatched rather than merging on a guess. A wrong match mixes two people's history and permissions; an unmatched account only waits for better evidence.
When something is deleted, is everything built from it deleted too?
Yes. A deletion at the source, or a request to erase material, carries through to everything built from that material: the facts taken from it, the text pieces, the search copies (embeddings) and the links and stored copies made from it. The search copies are deleted with their source, not kept behind. Backups expire on the published schedule, so deleted content leaves backups on that schedule instead of staying in them without a limit.
Does customer data train models, and which providers process it?
Customer data never trains any model. That rule is built into how Jilu works: customer data has no route into training, so the promise does not rest on a setting someone could change. For model providers, zero retention and no training are requirements Jilu places in its provider agreements. Pulling facts out and giving answers do not depend on one model provider, so Jilu can change providers without rebuilding the product. The sub-processor list will name the providers before any of them touches customer data, and customers can require processing to stay in a named region.
How will agents access the record over MCP?
Outside agents can reach the record over MCP, with the permissions of the person using them and no wider rights of their own. Reading comes before writing. Anything an agent writes passes through the same draft and review gate as everything else, so an agent cannot add to the record without a person's approval. Every agent action is logged against that person, so an outside agent is never an unnamed reader of the record. This connection is a design commitment under technical review, not a shipped feature.
How does Jilu measure answer accuracy?
Jilu's answers are tested, not taken on trust. The design checks three things: whether the source named for a claim really supports it; how much of the record carries sources; and how old the facts behind an answer are. A way to report a wrong answer sends failures into the review queue, where a person fixes the record itself, so a reported error is repaired at the source instead of repeating.
Can third-party skills or plugins be installed into the agent?
No. When Jilu launches, the agent uses only tools Jilu built. No tool, add-on or extension from anyone else can be installed into it, by an employee or by an admin, so no outside code runs inside the agent or reaches workspace data through it. The tools the agent has are listed for workspace admins, and an admin can switch one off for the workspace at any time, with effect at once.
Can I connect my own agent to Jilu?
Only if a workspace admin has approved and registered it. A personal agent connects under its owner's name, acts only with that person's permissions, and holds no access of its own. Every action it takes is logged against the person who owns it. An admin can revoke one registered agent at any time, without stopping other agents or the workspace agent, and the revocation takes effect at once. An unregistered personal agent cannot connect.
Security & privacy
The security page sets out the full programme: how Jilu protects the data, controls access, detects and responds, and proves each claim. The group after this one, For security teams, answers the questions a CISO review asks first.
How is company data secured?
Jilu is being built to be secure from the start. Permissions, access, retention, deletion and incident response are requirements of the product, not extras added later. Before a pilot connects company data, the customer should review the written architecture, data flow and controls.
Jilu will only claim a control or certification after it has been completed and independently verified.
Does Jilu respect the permissions in source tools?
Yes. Jilu only shows a person content they could already see in the source tool. If someone cannot open a Slack channel, Drive file or other source, Jilu must not reveal content from it. If their source access is removed, that change must carry through to Jilu.
We will test and write down how each connection carries over the source's permissions, before that connection is used for real.
What is the certification roadmap?
No certification is claimed today. The plan is SOC 2 Type I in year one, followed by SOC 2 Type II after the controls have been observed in operation. Type I checks that the controls are designed and in place; Type II checks that they keep working over time.
ISO 27001 is planned as international sales begin. The security page and customer documents will list only certifications that have actually been earned.
Will Jilu have independent security testing?
Yes. The plan is an independent penetration test every year once the production service is operating, with findings tracked through to a fix. Customers should be able to review the latest report or a clear summary under appropriate confidentiality terms.
What about GDPR, CCPA and customer DPAs?
Jilu is being built to meet GDPR and CCPA requirements. Before production use, customers will receive a data processing agreement that covers how data is used, subprocessors, retention, deletion and requests from people whose data is involved.
Jilu will not claim compliance until the legal terms and day-to-day operating processes are reviewed and in place.
Is customer content used to train a shared model?
Jilu does not use customer content to train models, and no shared model is trained on customer content. For model providers, zero retention and no training are requirements Jilu places in its provider agreements. That position needs to appear in the signed data terms, along with the subprocessors involved and any limited use needed to operate the service. Do not rely on a marketing page for this point. Check the contract.
Can sensitive sources be excluded?
They should be. Administrators need source-level control over what is connected, and teams need a clear path to exclude or remove material that does not belong in the archive. We confirm the available controls during setup.
For security teams
What limits the damage if Jilu is breached?
Jilu holds the record of how a company works, so a breach here matters more than at a tool that holds one kind of file. The design limits what any one breach can reach. Workspaces share one database (Postgres), but each workspace's data is sealed off from the others: every search runs under that workspace's own rules, so no search can cross into another workspace's record. Data is encrypted while it moves and while it is stored. Separate encryption keys for each company are on the roadmap, so one company's keys never unlock another company's record.
What Jilu collects sets the outer limit. The customer chooses which channels, spaces and kinds of data Jilu collects, and can leave any of them out entirely. Direct messages and HR spaces stay out unless the customer deliberately includes them. Meetings follow the same rules: one-to-ones and HR meetings stay out by default, and a meeting joins the record only when the customer has included meetings.
Who at Jilu can read customer data?
No one by default. Jilu staff hold no standing access to customer data. When support needs access to fix a problem, Jilu grants it only then, for a named reason, for a limited time, and only after the customer's workspace admin approves it. Each grant is written to a log the customer can see.
Contractors and outside reviewers never read customer data. Jilu staff do all support and review work under the same just-in-time rules: approved, time-boxed and logged. If a task cannot meet those rules, it does not happen.
A customer can ask who holds access to its data and see the record.
How does access into Jilu itself work?
Customers control access through their own identity provider. People sign in with SSO through that provider (Business and Enterprise), and MFA is enforced by the provider, not by a separate Jilu password. SCIM (Enterprise), the standard system for creating and removing accounts automatically, creates accounts when people join; when the account at the provider ends, the person's access to Jilu ends with it.
Separate admin roles keep everyday use apart from administration, and each role holds only the access its work needs. Changing what Jilu collects is an admin action. Jilu logs every such change and makes the log visible to the customer. The audit log the customer can see is included on Business and Enterprise.
What does the audit log cover, and can we export it?
The audit log only ever grows; no entry can be edited or removed. Each entry carries a fingerprint (a hash) of the entry before it, and the latest fingerprint is stored outside the database every hour and checked every day. Every read writes to the log, whether a person or the agent did the reading. It records who or what read the data, what it read, and when, including the agent's own actions. Change a written entry, and the chain of fingerprints breaks, and the daily check catches it.
Logs can be exported, so a customer's security team can pull Jilu events into its own monitoring and review them beside its other systems. Audit events are kept for the life of the workspace and for one year after it closes, then deleted. How long they are kept can be set for each workspace.
How does Jilu respond to a security incident?
The data processing agreement will carry the notification commitment. Jilu will notify a customer within 48 hours of confirming an incident that touches that customer's data. The notice states what was affected, what Jilu has done, and the name of the person to contact.
During an investigation, a workspace admin can pause collection, and collection stays paused until the customer restarts it.
Can we pause or stop the agent?
Yes. A workspace admin can pause collection or stop the agent at any time. The stop takes effect at once: the agent stops answering questions and stops drafting records until an admin starts it again. Boards stay readable while the agent is stopped. Every pause and restart is written to the audit log.
Which other companies handle our data, and what do model providers do with it?
Jilu will publish its sub-processor list before any provider touches customer data, with the region each provider works in and a change-notice window, so a customer can review a new provider before it touches data. Model providers will sit on that list.
Keeping nothing and training nothing are requirements Jilu will place in every provider agreement: the provider keeps no data after processing it, and never trains a model on customer data.
What evidence can we request for a security review?
A security team can request the security programme summary and the certification roadmap today; both are available on request. The data processing agreement and the named sub-processor list are not final documents yet, and Jilu says so rather than presenting them as ready. Penetration test summaries are available under NDA when they exist.
Jilu claims no certification it has not earned, and the review pack states the current position plainly. Security questions route through the security page.
How does exit work, including deletion and legal hold?
A customer can stop Jilu collecting at any time. The customer then exports the full record in open formats, so the archive stays readable without Jilu. Jilu then deletes the customer's live data and everything built from it, and the deletion is logged. Backups are emptied within 35 days.
If the customer places a legal hold, deletion pauses for the held material only. The hold belongs to the customer: Jilu holds the material while the hold stands and deletes it when the customer lifts the hold.
Do model providers keep workspace data for abuse monitoring?
No, and a workspace can check that for itself. Jilu requires zero-retention terms from every model provider it uses, and that requirement covers abuse monitoring as well as training. If a provider's terms include an abuse monitoring hold that Jilu cannot remove, the hold and its length are named on the sub-processor list before any workspace is connected to that provider, and a workspace admin can see which providers hold what, and for how long, at any time.
How does a workspace revoke an external agent's access?
Each external agent a workspace admits is registered by name, tied to that workspace and to the people who may invoke it. An admin can revoke one agent's registration on its own, without affecting the people who used it or any other agent. Revocation takes effect at once: Jilu checks permissions when a question is asked, so the revoked agent's next request is refused, and the revocation is written to the audit log with the admin who made it.
What happens to agent access when an employee leaves?
It ends with the account. When a person is deprovisioned through SCIM or the identity provider, the agent registrations in their name end at the same time, and the tokens and sessions tied to those registrations are revoked. No agent keeps access to the record through a person who has left. Admins can also revoke a single agent's registration at any time while its owner is still in the workspace.
What happens if a password or API key ends up in a source Jilu collects?
Jilu checks sources as they are collected for material that looks like a credential, such as a password, an API key or a token, and flags anything it finds. A flagged passage is shown to the reviewer before publication, so a credential is not carried into the record unnoticed. When a reviewer or an admin confirms that an item is a real secret, Jilu deletes it, and the deletion carries through to everything built from it, including drafts, summaries and answers built from it, as every deletion does. The flag, the review decision and the deletion are written to the audit log.
For employees and admins
Is Jilu employee surveillance?
No. Jilu keeps the company's record of decisions and how the work works. It does not score individuals. It does not analyze sentiment or tone. It does not track activity, keystrokes or time, and it produces no per-person productivity data for managers.
These are product boundaries, not settings an admin can switch on later: Jilu does not collect the data those features would need. Workplace emotion recognition is also prohibited under the EU AI Act, and Jilu stays on the right side of that line by design, because it never attempts it.
Does Jilu scan files on employees' computers?
No. Jilu does not scan local drives, and it does not collect files that sit outside a connected source. Jilu collects only from the sources a workspace admin chooses, through the connected tool for that source. An admin sees what a source would bring in before it goes live, can remove a source at any time, and can pause collection. Content in a source still follows the permissions of that source whenever a question is asked.
Can my manager see what I ask the agent?
No. A question belongs to the person who asked it. The audit log records that the agent acted, and a security team can review agent actions for safety. Jilu does not build a feed of what individual employees asked or read, and it gives managers no view into anyone's questions.
What can the agent do on its own?
The agent answers questions and drafts knowledge from the sources a workspace has chosen. It works with the permissions of the person who asks, so it cannot reach anything that person cannot reach. It cannot publish knowledge on its own. Every draft waits for a person to review and publish it. Actions with consequences beyond drafting require the approval of a person first, and Jilu logs each agent action against the person who delegated it. Admins choose which sources Jilu may use, and a source left out stays out of the record.
Our permissions in Slack and Drive are a mess. What happens on day one?
Nothing connects blind. Before a source goes live, admins first see what would be connected: the channels, spaces and folders it would include, and they can exclude any of them. Rollout runs team by team, not company-wide on day one.
Jilu checks access when a question is asked, against the source's permissions as they stand now, so a fix in Slack or Drive takes effect in Jilu at the same time. Material shared too widely is a problem in the source, and any search tool would show it. That first look exists so the company sees the material before employees do.
What do employees get told when Jilu is rolled out?
Every rollout includes a plain employee notice. It states what Jilu collects, why it collects it, what stays out, and what Jilu never does. Employees can read the record Jilu builds about their work, through the boards. No one gets recorded into a system they cannot open.
How long are employees' questions and the agent's answers kept, and are they covered by legal hold?
Questions and answers are part of the workspace record. They are kept under the retention policy the workspace admin sets, the same policy that governs the rest of the record, and they are included when a workspace exports its record. If a customer places a legal hold on its record, questions, answers and the audit entries tied to them are preserved under that hold until the customer releases it. When the retention period ends, or an admin deletes them, the deletion reaches everything built from them, as it does for the rest of the record.
Can Jilu speak in a conversation, or only answer when it is asked?
Jilu can speak up in a conversation it is part of, in Slack, in Teams and in Jilu's own surfaces. It does so in four cases: a decision is being made with no reasons given; something said conflicts with a published record, in which case it says so and names the source; a commitment is made with no owner; or a question is asked that the record already answers.
It speaks rarely and briefly. Everything it says is logged. Nothing it says changes the record by itself: what it proposes enters as a draft and a person approves it. It only uses material the people in that conversation are allowed to see. A workspace admin can limit Jilu to answering when called on, choose where it may speak, or pause it at once.
Are conversations kept as part of the record?
A conversation Jilu is part of is kept, and it can be searched and cited as a source later, on the same footing as a document or a decision record. What a conversation settles reaches the published record through the same draft, review and publish gate as everything else.
Collection is opt-in for each source, as with every Jilu source. Sensitive conversations stay out of scope by default, including one-to-ones and HR conversations. A workspace admin can pause collection at any time, with boards staying readable and every pause and restart written to the audit log.
Legal & ownership
Who owns the company's data and archive?
The company should retain ownership of its source content and the archive created from it. Jilu should receive only the limited rights needed to run the service. The binding answer will be in the customer agreement and data-processing terms supplied before access.
Does Jilu sell customer data or use it for other customers?
No. The customer company owns its source content and the archive Jilu creates from it. Jilu holds only the limited rights it needs to run the service for that company. Jilu does not sell customer data, does not provide customer data to data brokers, and does not use one company's record to train a model or to serve another company. For model providers, zero retention and no training are requirements Jilu places in its provider agreements, and Jilu staff hold no standing access to customer data.
What legal documents are available?
Production use should include a customer agreement, privacy terms, data-processing terms, a subprocessor list and security documentation. Jilu supplies these documents before access, and teams review them before connecting company data.
Can retention and deletion follow company policy?
They need to. Different records have different legal and day-to-day requirements. Retention periods, deletion requests and legal holds should be agreed in writing rather than hidden in a setting nobody can change.
What happens to the archive if a company stops using Jilu?
A permanent record is only trustworthy if it is portable. Jilu's intended offboarding standard is an export of the archive and its source references, followed by deletion after the agreed handover period. The exact format, timing and backup-deletion terms must be written into the contract before a company starts.
Sales & access
Who is Jilu for?
Teams whose important decisions are spread across messages, documents, meetings and project tools, especially growing companies where people change roles and the same questions keep returning.
What comes with the Team plan?
$24 per seat per month billed annually, or $29 month-to-month. Add the paid seats your team needs. Team includes shared knowledge boards, the agent that answers from the record with sources, the main connections, standard workspace permissions and email support.
It is designed for one team establishing a shared company memory.
What comes with the Business plan?
$39 per seat per month billed annually, or $47 month-to-month. Business is priced per paid seat and includes everything in Team, plus a tool that builds the company wiki, all connections, single sign-on, an audit log, wider admin controls and priority support.
It is designed for organizations bringing multiple teams into one shared workspace.
What comes with the Enterprise plan?
Custom pricing and seat commitments. Enterprise is built for company-wide rollouts. It includes everything in Business, plus SCIM, data residency options, a DPA, a tailored security and implementation review, and dedicated support.
Deployment requirements, support coverage and any additional controls are agreed with the customer.
What does the 90-day pilot look like?
90 days, scoped to one team with up to 50 seats. The team receives full platform access for the agreed scope. At kickoff, we choose the sources and workflow to evaluate, confirm permissions, and agree on measurable success criteria.
At the end of the 90 days, we review the results together. If the company continues, the boards, the record and the permissions carry over unchanged.
The Jilu 90-day pilot page sets out what the team gets, how the 90 days run, and what the team keeps at the end.
When will Jilu launch?
Teams start now, through the 90-day pilot: request access, and Jilu sets up the workspace with your first team. A public launch date has not been announced.
The pilot runs on the full platform for the agreed scope. If the company continues after the end-of-pilot review, the published per-seat pricing applies, and the boards, the record and the permissions carry over unchanged.
Teams can request access now and say what they would like Jilu to make possible.
What does requesting access include?
A review, limited to the job you have in mind: the sources, the permissions, the security needs and what success looks like, all before any data is connected. Each team starts with the 90-day pilot, and Jilu sets up the workspace with the team.
Can we evaluate Jilu in a sandbox?
Yes. During the technical review, Jilu will provide a test space: a separate place where your team can try Jilu with sample information, or real information with the sensitive parts blacked out, before Jilu is used for real work. We agree which connections are ready, an example of your work to test with, and the limits on data, as part of the pilot. That way the test reflects your company without putting its live systems at risk.
How does a Jilu sandbox work?
A sandbox is a separate Jilu workspace, set apart from your company's real workspace, where your team can test Jilu before trusting it with real work. You can add sample work or connect a test source and run the whole flow end to end: watch sources become drafts, review and approve them, build the wiki, and ask questions and see the answers arrive with their sources.
Nothing in the sandbox touches your company's live record. The two workspaces are kept apart, the same way any two workspaces are, and the same permissions and the same draft, review and approve gate apply inside the sandbox. When your team is ready, the real workspace starts clean, and the sandbox can be deleted, including everything in it.
Can we see a demo first?
Yes. Request access and describe the decisions you want to keep. A useful demo follows a real example of your team's work, without asking you to show live company data.
Design partners
What is the design partner programme?
Jilu is being built with a small number of design partners: companies that work with Jilu while the product is finished. The programme is where the product is shaped with the customers who will use it. Partners work with the product while it is being built, and what they find changes what gets built.
Within the programme, Jilu is customised to how the partner company actually works: its teams, its sources and the record it keeps. That customisation is part of the programme work. It is not bespoke development on demand.
Who is the programme for?
Companies whose decisions and know-how get lost in the ordinary run of work, and that are willing to test a product that is still being finished and say plainly what holds up and what does not.
What does a design partner do?
A design partner brings one team. The team shares the decisions and work that get lost in its ordinary run of things, and tests Jilu on that work. The partner and Jilu talk regularly, and the partner says plainly what holds up and what does not.
How do the 90 days run?
The programme runs as a 90-day pilot, scoped to one team of up to 50 seats. The team has full access to the platform for the agreed scope, for the 90 days. At the start, the partner and Jilu agree the sources and the work to test, confirm permissions, and agree what success means.
At the end of the 90 days, the partner and Jilu review the results together, against what was agreed at the start.
What happens to the record afterwards?
The team keeps its record either way. If the company continues with Jilu, the boards, the record and the permissions carry over unchanged. If it does not continue, the company exports its full record in open formats, and Jilu then deletes the company's data.
How does a company become a design partner?
Request access and say that you are interested in the design partner programme. Jilu talks with the company about the team and the work it would test, and the two sides put a short agreement in place before any data is connected.
Nonprofits
Why does a nonprofit need a record like Jilu's?
In a nonprofit, much of what the organization knows is held by one or two people: why a program runs the way it does, what a funder was told, and how a policy came to be. Boards, funders and auditors ask for decisions and the reasons behind them, often months after the decision was made. Jilu keeps each decision with its reasons, its sources and the person who approved it, in one record the organization can open and check when those questions come.
What changes for a small nonprofit?
A small nonprofit runs on part-time staff, volunteers and short contracts, so people arrive and leave often. When the person who knew how something worked leaves, the decision and its reasons leave with them. A small team has no spare capacity to rebuild that knowledge each time. When the record is kept in Jilu, the decision and its reasons stay with the organization after the person has gone.
How is Jilu used at a nonprofit?
Decisions are kept with their reasons, their sources and the person who approved them. A new staff member or volunteer can ask a question and get the answer with its sources, instead of reconstructing how something was decided.
Handovers stop depending on one person's memory and inbox, because the record holds what was decided and why. Questions from a board or a funder can be answered from the record, with the sources behind the answer visible.
How does a nonprofit start?
A nonprofit can evaluate Jilu in a sandbox first: a separate test workspace where the team can run the full flow on sample work, kept apart from the organization's live record. A nonprofit can also ask about the design partner programme, where a small number of organizations work with Jilu while the product is finished.