JiCoRedact. Approve. Document.
Frequently asked questions
What JiCo does, what it does not do, and how you can check both.
What JiCo is
What exactly does JiCo do?
JiCo sits as a local process between your AI assistant and Jira or Confluence. On the way to the AI model it replaces personal data with placeholders; on the way back it resolves them again — locally, on your machine. Every changing action is held and presented to a human: command, affected object, effect. What went out, and who approved what, is written to an append-only log.
Is this a cloud, a service, a subscription?
No. JiCo runs as a program on the machine that also runs your AI client — .pkg for macOS, .msi for Windows. Licensing is per installation and checked offline; you pay per term, without automatic renewal. A shared multi-tenant installation is explicitly not intended.
Which AI model does JiCo use?
The AI model of your AI client, for example Claude — it only receives what JiCo has redacted beforehand. JiCo brings no chat window of its own: it is the gateway, not the assistant. A local AI model (Ollama) is needed for two things only, both of which have to stay local — checking image attachments and building the knowledge search; a cloud AI model is refused for either.
Limits — what we would rather say here than in the small print
Does JiCo detect all personal data?
No, and no detection does. Reliably detected are check-digit numbers (social security, IBAN, credit card), secrets, and structural entities from Jira itself — assignee, reporter, comment authors, mentions. Free text is statistical: JiCo finds a lot there, but not everything. That is why the redaction balance is shown and not hidden.
Does JiCo prevent every data leak?
No. JiCo protects what goes through JiCo. The write path is sealed: the AI can only change things through JiCo, and only after approval. The read path depends on a clean client configuration — if someone connects a second, unprotected Jira access next to it, the AI takes the data from there and JiCo never learns of it.
Does JiCo make us GDPR compliant?
Compliant describes a processing activity, not a tool. JiCo supports your obligations: data minimisation on the way to the AI model, evidence of who approved what, and a log that an auditor can read. The assessment stays with you.
Is JiCo certified or audited?
No. There is a test programme in the CISO guide, which you can request via support@getjico.com and with which you can verify the promises yourself — that is something different from a certificate, and we call it what it is.
Is the pseudonymisation anonymous?
No, it is deliberately reversible — otherwise the AI could not write a usable comment. The mapping stays local, encrypted and time-limited; the AI model only ever sees the placeholder, the human in the approval window sees the plain text.
Operations
What happens if the ticket changes between proposal and approval?
Then it is not executed. Before every execution JiCo reads the state again; if the object has been changed or commented since the analysis, the action moves to the follow-up list with a reason. Even a change to a field the action does not touch blocks it.
Does JiCo work when Jira is unreachable?
Reading yes, from the local redacted store — marked "not verified live". Approvals can be queued and are executed as soon as Jira answers again, and only on an unchanged state.
Which Jira edition is supported?
Jira Data Center is live. Confluence Data Center speaks the same protocol, and both have been verified against a real instance: reading (spaces, a page with title, body, version, structural people) and commenting through the approval gate — the comment appeared only after approval, and the page version stayed unchanged. That run surfaced three defects, all fixed. Jira Cloud and Confluence Cloud are built as well, for reading and writing, and verified against a real Cloud instance; there you sign in with an e-mail address and API token.
How do I use Jira Data Center with Claude?
Through an MCP server that can reach your Jira — JiCo is one, running locally on your computer. You connect Claude Desktop or Claude Code in the Approval Cockpit with one click; personal data is redacted before Claude sees it, and every write action waits for your approval. The steps are on the page Jira Data Center with Claude.
Where does our data sit?
On your machine. The placeholder ↔ plain text mapping is encrypted and expires after 96 hours at the latest; the redacted cache holds redacted text only. Only what you configured or triggered goes outside: Jira or Confluence, the AI model and — on your click — the customer portal for licence, update or remote access. JiCo blocks everything else.
Is the software available in English?
Not yet. This website is; the program's own texts, commands and the approval panel are German. Tell us if that blocks a deployment — it is on the list, and demand decides the order.
Where the status is stated
How do I know what is actually built?
On the Features page: it lists only what is built. It is generated from the feature list, and an entry appears only once its feature is complete.