JiCoRedact. Approve. Document.
Work in Jira & Confluence with Claude & co. – without disclosing personal data.
Personal data is redacted before an AI model sees it. JiCo sits locally between your AI assistant and Jira or Confluence and requires human approval for every write action.
At a glance
- Redact: personal data becomes placeholders before the AI model sees it.
- Approve: every write action waits for a human — presented in business language.
- Evidence: proposal, decision and execution are logged without gaps, with no plain-text personal data.
- Operation: runs locally on your computer, with Claude or fully locally with Ollama — try it free for 7 days.
The problem
An AI assistant on Jira is set up in ten minutes — and in the same ten minutes it has handed names, addresses, social security numbers and half a personnel file from your tickets to someone else's AI model. What has been in an AI model's context, no delete key brings back.
At the same time that assistant is allowed to write: change tickets, post comments, overwrite fields — without anyone having seen beforehand what exactly it does and what that sets off.
The obvious answers don't hold. A filter downstream arrives too late. A ban gives away the benefit. A prompt that asks for restraint is not a control, it is a request.
Three gates in the data path
PII gate — redaction happens before the AI model
People become placeholders. Numbers with a check digit — social security, IBAN, credit card — and secrets are masked, or they hold the whole unit back. Attachments that cannot be checked locally never leave at all. The AI model sees «Person 1» — depending on the setting also initials such as «@PM» —, never the full name, never the Jira ID.
Approval gate — human in the loop: no write access without a human
Every changing action is held and presented in plain professional language: which object, which effect, what it makes obsolete. Unknown tools count as writing — the classification is fail-closed. And an approval only holds for the state it was based on: if the ticket has been updated or commented since, nothing is executed.
Audit trail — evidenced, not asserted
Proposal, decision and execution are recorded append-only and correlated per command. Without plain-text personal data, only with the redaction balance: how many matches of which type were redacted.
What a case looks like
One ticket, as the AI model receives it — and as the approval desk presents it. The redacted part never left the machine.
What the ticket says
Reporter: Sabine Herzog — billing for policy 756.1234.5678.97 has been failing since the update. Please call back on +41 79 123 45 67, or write to s.herzog@example.ch.
What the AI model sees of it
«Person 1», ###MASKIERT###, «Telefon 1», «E-Mail 1» — three placeholders and one mask, no plaintext, not even the Jira ID behind them.
What you are shown
Command: add a comment to ORBIT-4711 · Effect: one new comment, no field change · State: re-read two minutes ago, unchanged since the analysis.
What it gives you day to day
- Assignments instead of clicking. JiCo works through tasks one by one, as a queue or on a schedule. The reading part runs without you; at the end an approval folder is waiting.
- Grouped by topic and sanity-checked. Findings are condensed; contradictions, duplicates and scope deviations surface before the approval rather than after it.
- Less AI slop. A project glossary with boundaries — "a defect is not a change request" — and house conventions with their follow-up duties go into the AI model context. Thought two steps ahead.
- Follow-up list. What a change makes obsolete in earlier statements is listed with it.
Where your data sits
JiCo runs on your machine. You work with the AI of your choice, for example Claude. It only ever sees your tickets with personal data redacted. Separately, you approve in the approval desk what the AI wants to change. Without your approval, it changes nothing.
Deployment options
Local — macOS and Windows
Installer from the customer portal for the individual workstation: .pkg for macOS, .msi for Windows, same range of functions. Approval Cockpit in the browser, icon in the menu bar (macOS) or in the notification area of the taskbar (Windows), connection to Claude Desktop, Claude Code or GitHub Copilot in VS Code.
Team setup — intranet / self-hosted, on request
For teams on the same network: self-operated on a Mac on site or as a dedicated JiCo appliance. All data stays in your network, no multi-tenant SaaS. This is our target architecture for several users at one site; today it is set up individually with you, not as a self-service download. Request team setup
Hosted — a dedicated Mac per customer, coming soon
For teams without their own hardware: a dedicated Mac mini just for your JiCo instance, in a data centre in the EU (Paris) or in Switzerland — no shared server, no multi-tenant SaaS.
Enterprise setup — on request
For larger organisations: a dedicated instance in your own data centre or your own private cloud, with your requirements for operations, roles and permissions. Planned individually with you. Request enterprise setup
Where JiCo stands in the market
No product combines what JiCo does. Three things make the difference:
- Redacted before the AI model reads it. Personal data becomes placeholders before the AI model sees it — not filtered afterwards.
- No write access without your approval. Every action waits for a person. If the ticket has been updated or commented on since, nothing is executed.
- For Jira Data Center and Jira Cloud, running locally. JiCo runs on your machine, with an offline licence and no multi-tenant SaaS.
See the full comparison → — Atlassian Rovo, Strac, Portkey, Kong and IBM watsonx.governance, criterion by criterion, with sources.
Pricing
JiCo costs €490 (CHF 460.60) per month and installation. As an introductory price, orders placed by 31 December 2026 pay €295 (CHF 277.30) — for the whole term purchased. From 1 to 200 installations — net plus VAT, approval for every write action included.
| Contingent tier | Installations | Price per month | Price per month (CHF) |
|---|---|---|---|
| Professional | 1–200 | €295 until 31 Dec 2026, then €490 | CHF 277.30 until 31 Dec 2026, then CHF 460.60 |
| Enterprise | over 200 | discount schedule on request | discount schedule on request |
Show details
- Every licence includes Jira and Confluence (Basis) — no module choice.
- One licence for reading and writing — every write action needs approval; no tiers.
- Term of 30 or 365 days; the 365-day term gets an additional −10% annual discount (otherwise the monthly price times 12).
- Invoice in EUR or CHF, as you choose; exchange rate 1 € = CHF 0.94 (rounded, as of 2026-09-21).
- Try it free: register in the customer portal — 7 days, reading and writing, at most 3 executed write actions per day, no payment details, no automatic transition into a contract.
- Buy in the customer portal: 1 to 200 installations, 30 or 365 days, on invoice, payable by bank transfer, card or SEPA, without automatic renewal.
- Enterprise (over 200 installations): discount schedule on request, no list price. Request pricing information
Getting started
JiCo is a local MCP server: install it, tick the Jira projects and request the licence during setup, enter it in your AI client.
- Quick start — from sign-in to your first question, in six steps
- Customer portal: register, buy, download the package — sign in by link to your email address
- Frequently asked questions — what JiCo does, what it does not, and how you check both
- Blog — background and decisions, as they were made
- User manual, deployment and operations guides are in the customer portal after sign-in
- Architecture and SBOM on request, for CISO briefings: request documents
Stay up to date
New detection layers, connectors and decisions — occasionally by email, never more than that. The same applies if you are already a customer: signing up here is independent of licence and credentials, and you can withdraw at any time via the unsubscribe link in every email.
Or follow JiCo on LinkedIn · WhatsApp (channel)