The Kanon web app is the reading and management surface for your knowledge base. You use the plugin to generate content; you use the web app to read, review, and act on it.

Pages

Knowledge base (`/knowledge-base/{repo-id}`)

The main reading surface. A sidebar shows spaces → domains → features. Each scanned feature links to its guide page.

Overview page: totals (domains, features, claims, facts), coverage bar, honest gaps (unscanned features, empty domains), domain table.

Feature guide page: the full generated guide with all sections — narrative, principles, lifecycle, business risks, vendor dependencies, user journey, known limitations, compliance flags, data model, route map, jobs, flags, events, experiments, test plan, and glossary. Every behavioral sentence carries a trust badge (verified/asserted/drifted) and a citation linking back to the source code.

Tracking plan (`/events/{repo-id}`)

Manage your analytics event inventory. The tracking plan reconciles what's in the code against what SHOULD be tracked.

Statuses: implemented (in code, aligned), suggested (Kanon recommends adding), approved (user accepted, not yet coded), missing (in plan, removed from code), misaligned (in code but differs from plan).

Actions: approve/reject suggestions, add events manually, edit existing entries, create a Linear ticket for the diff between the plan and the code.

Flag plan (`/flags/{repo-id}`)

Feature flag lifecycle management. Tracks every flag with its provider, lifecycle (temporary/permanent/experiment/ops/kill-switch), owner, cleanup date, and gate site count.

Statuses: implemented, planned, stale (marked for cleanup), missing, undocumented (in code but not in the plan).

Actions: document undocumented flags, mark stale, set cleanup dates, create a Linear ticket for maintenance.

Experiments (`/experiments/{repo-id}`)

A/B test inventory. Shows every experiment detected in the code with its provider, variants, hypothesis, metric, and status (active/concluded).

Concluded experiments get a warning callout — they should be cleaned up (ship the winner, remove the branching code).

Security audit (`/security/{repo-id}`)

Static analysis findings grouped by OWASP category. Each finding shows severity (critical/high/medium/low), description, remediation guidance, code evidence, and source location.

Not a penetration test — pattern-based detection of common anti-patterns (hardcoded secrets, SQL injection, XSS, broken auth, etc.).

Testing plan (`/testing/{repo-id}`)

E2E test scenarios across the whole product. Grouped by feature with priority (critical/high/normal/low), step counts, category (happy path/error path/edge case), and automation status.

Review taxonomy (`/knowledge-base/{repo-id}/review`)

The approval UI for discovery proposals. Accept, edit & accept, or reject each proposed domain and feature. Bulk "Accept all" per domain.

Changes (`/knowledge-base/{repo-id}/changes`)

Pending product changes — spec changes from Linear tickets, scan-detected drift, or GitHub PRs. Shows status (open/implementing/resolved/dismissed) and affected claims.

Each open change offers two execution paths: copy the /kanon:work command for a terminal session, or — when a worker daemon is online — click Run with worker to queue it. A queued change shows a live status chip linking to its session view (/knowledge-base/{repo-id}/tasks/{task-id}): a step-by-step trace of what the worker is doing, a cancel button, and the PR link when it lands.

Settings (`/settings`)

Members — invite users, manage roles (admin/member/viewer)
Integrations — connect Linear (OAuth, team binding per repo), connect GitHub
API tokens — machine access tokens for CI and the plugin

AI chat

Every knowledge base page has an AI chat assistant. Ask questions about your product and get grounded answers:

Answers cite specific claims and sections from the knowledge base
Tables and diagrams are rendered as widgets, not inline markdown
When you discuss changing product behavior, the AI proactively drafts a Linear ticket (categorized as bug/improvement/feature with a comprehensive template)
When you ask a factual question, it answers without suggesting changes

Web app overview