Skip to content

Changelog

All notable changes to AppControl will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[1.24.5]

  • Use component roles, documented routes and startup prerequisites when generating a map. Distinguish links that describe traffic from dependencies that control startup and recovery.
  • Refine an existing draft against its source revision, the component catalogue and configured sites. Apply selected changes to the draft before deciding whether to publish.
  • Add an optional instruction to focus the analysis. Review recommendations on operational commands and procedures alongside the evidence cited by the model.
  • Preselect the best supported catalogue match using product and version evidence. The graph marks tentative selections and lets operators confirm or replace them before publication.
  • Look for dependencies in connection settings and procedures. Review can propose inferred links as weak dependencies for the operator to verify.
  • Choose the execution sites and agents of unassigned components before individual publication. Updates retain existing assignments and include new components in placement review.
  • Remove the prompt contract from Swagger. Validation and connector diagnostics remain documented.

[1.24.4]

  • Review suggested maps on a wider, zoomable graph with full-screen mode and the X4B theme. Edit the draft before publication and compare it with the published map when applying an update.
  • Assign or confirm catalogue entries directly from the review. Matching uses explicit product evidence and aliases from source documents. The graph marks components that still need a type and displays documented obsolescence. Reviewed components without a catalogue type can be published as Custom.
  • Show the user's available applications in the embedded X4B header when the host provides the session's module list. Confirm sign-out in both the native and embedded headers.
  • Install, update or uninstall from the Windows OpenShift kit with Gerer-AppControl.ps1. In-place updates reuse the saved installation and preserve manually managed network policies, including those used for internal document services.
  • Run release qualification as the required platform artifacts become available, with targeted retries for transient Windows startup and connection errors.

[1.24.3]

  • Browse and edit the component catalogue from X4B, use it when creating components, and see catalogue labels while reviewing an import. Map generation provides the catalogue and site references to the AI, shows uncertain matches for review, and suggests similar site names.
  • Edit generated topology proposals as drafts, rerun generation on demand, and compare with the published map before applying an update. Publication keeps the application identity, existing component identities, agents and commands. If the source or the published topology changed after review, reopen the comparison before publishing. Remove existing components explicitly in the map editor.
  • Correct an external application's URL or callback from X4B without registering the application again. Redirect embedded users to the X4B login page after successful logout.
  • Configure, test, suspend and reactivate the document connector as an X4B administrator. The settings screen is available by default for public HTTPS and for same-namespace Services on OpenShift. Deployment-specific allowlists and existing connector Secrets are respected.
  • Prepare a scoped OpenShift egress policy for internal HTTP with the Windows kit. Upgrades retain connector settings, Secret references and the dedicated policy. Fix the OpenShift command context used when configuring the connector.

[1.24.2]

  • Add deployment-only connector options for HTTP to an explicitly named internal Service and for temporarily accepting untrusted HTTPS certificates in development.
  • Log connector transport failures and HTTP statuses without logging tokens or response bodies.
  • Allow the OpenShift installer to resume configuration after demo verification fails, and the uninstaller to clean up installed simulation workloads before removing AppControl.
  • Add connector settings and connection tests to X4B, and improve catalogue and site matching during map generation.

[1.24.1]

The v1.24.0 candidate remained unpublished after a native sign-in issue was identified.

  • Correct the X4B sign-in flow for native AppControl pages.
  • Introduce the X4B application launcher, shared navigation and administration of local accounts and module access. Registered external hosts can resume a deep link after signing in with X4B.
  • Expose a read-only AppControl status badge and isolated simulation to registered hosts, under the current user's map and site permissions.
  • Move map generation into its own X4B workspace. Match source components against the full technical catalogue, accept aliases from customer exports and flag uncertain matches for review.
  • Import large catalogues through the Windows OpenShift kit in bounded requests.

[1.23.1]

The v1.23.0 candidate remained unpublished after the Azure gateway image scan reported CVE-2026-84243. This release updates that image, verifies image scans with comprehensive detection and rejects missing inventories and suppressed findings. All severities and vulnerabilities without a fix are included.

AppControl Embedded

  • Embed a live AppControl map in an approved documentation site with a Web Component. Each simulation uses a private copy of the map and dedicated execution agents. AppControl controls session scope, origin, expiry and audit history.
  • Explore the impact of a failure and simulate application recovery using the AppControl runtime. The main application distinguishes ordinary restart, application reconstruction and infrastructure reconstruction.

Map generation and catalogue

  • Add an X4B workspace that reads application keys and document hashes from a configured source, queues map generation through the customer's AI and validates each candidate. Operators can publish a first map or review the topology changes proposed for an existing map.
  • Preview and update technical catalogue entries in bulk. Catalogue commands prefill new components and can be adjusted per map. The generation contract can include a selection of catalogue entries.

OpenShift and certificate maintenance

  • Deploy a named instance with its sites, hosting, gateways, agents, teams, users and catalogue from one PowerShell configuration. Configure approved Embedded origins and dedicated simulation sites in the same file.
  • Show certificate and token expiry by component. The Windows connector processes approved renewals for installer-managed credentials and records their results. Certificate authorities, platform-managed HTTPS and external agents follow their own renewal procedures.
  • Keep the banking demonstration available from the lightweight kit. The installer can upgrade an existing instance with its original deployment name and private state.

[1.22.0] - 2026-11-06

  • Introduce X4B (XComponent for Business), the INVIVOO platform containing AppControl.
  • Bundle the official publisher logo, refresh sign-in and module navigation, and default to a persistent dark theme.
  • Include Windows installer improvements for explicit public URLs, administrator selection, post-install users, teams, sites, gateways and agents, and preserved registry secrets.
  • Preserve compatibility with existing AppControl API routes, image package names and application data.

[1.21.1] - 2026-11-03

The immutable 1.21.0 tag remained unpublished after the release dependency gate identified CVE-2026-118764 in development tooling. This delivery updates the ESLint toolchain and pins its compatible glob consumers to minimatch 11.0.1, without weakening the all-severity production/development release scan. It also updates benchmark fixtures for organization-scoped permission evaluation; runtime permission rules are unchanged. The release includes all features below.

AI-assisted map generation and publication

  • Provide a versioned generation contract with the map ontology, JSON Schema, catalogue selection, examples, Linux/Windows mock command templates and ETag caching. The customer's AI generates the JSON; AppControl makes no model call.
  • Add scoped publication context, prepare and import APIs with explicit placements by site, enrolled host or agent ID, and team grants. A single active application can span several sites. Existing map commands are preserved by default.
  • Treat simulated commands as a generation-prompt choice: the resulting maps use ordinary agents and the ordinary import workflow alongside real applications. Server-side mock materialization remains an explicit optional helper.
  • Add Swagger UI/OpenAPI documentation, PowerShell integration examples and a replayable multi-site demonstration with file-based start/check/stop commands.

Teams and access control

  • Manage application assignments and permissions from Teams in the UI.
  • Enforce organization, site, workspace, expiry and permission-ceiling checks; permission revocations take effect without requiring a new user session.
  • Validate destination bindings before profile activation and switchover, and prevent an import update from silently granting its editor Owner access.
  • Preserve organization-scoped 404 responses and reject expired share links.

Windows and OpenShift delivery

  • Include publication helpers and optional post-install site/agent provisioning in the lightweight PowerShell/oc kit, without bundled images or executables.
  • Retain targeted uninstall, explicit upgrades, Artifactory image rewriting and integrated/external PostgreSQL choices. External PostgreSQL still requires a pre-created dedicated database and an application account with schema rights; the installer does not accept a separate database administrator account.
  • Require an explicit image lock when assembling a new product version so the lightweight kit cannot silently reuse historical images.
  • Preserve demo scenarios in both source and corporate release assets. Release scans and runtime qualification refer to the actual delivered image digests.

Scope

  • Map publication is create-only; it is not an automatic update or an idempotent batch transaction. Review generated commands before import, because periodic checks may run as soon as components are bound to agents.
  • The OpenShift evaluation remains a single-instance deployment with evaluation PKI. Synthetic command tests do not qualify a customer's production procedures.

[1.20.4] - 2026-10-28

The 1.20.0 tag did not produce a release because GitHub rejected workflow permissions before any build. The 1.20.1 release gates then caught two cold-start installation defects. The 1.20.2 attempt was stopped after inspection of raw RustSec results identified eight advisories despite a successful job status. The 1.20.3 attempt was stopped to correct image references in the generated public documentation and deployment scripts. These tags remain immutable and unpublished; 1.20.4 includes the dependency updates, reliable failure propagation, validated public-registry substitutions and all changes below.

Security validation corrections

  • Update h2 to 0.4.16, quinn-proto to 0.11.18, rustls-webpki to 0.103.16 and webpki-roots to 1.0.10 to resolve the eight RustSec vulnerability advisories.
  • Resolve unsoundness advisories with anyhow 1.0.103, event-listener 5.4.2 and tokio 1.53.3; retain any remaining informational warnings in the audit report.
  • Propagate RustSec and test failures through log pipelines, verify the audit JSON explicitly, and retain audit reports even when the job fails.

Installation regression fixes

  • Align the non-root PostgreSQL image user with its Alpine runtime so new Docker data volumes initialize correctly; keep the existing data directory layout.
  • Fix optional-body argument handling in the demo bootstrap HTTP helper so fetching the enrollment CA sends exactly one authenticated request.
  • Start only the intended Compose test services, handle the packaged agent entrypoint correctly, and retain initialization logs on test failures.

Release and installation

  • Rewrite container registry ownership before repository branding, and reject invalid generated image names in the public documentation and scripts.
  • Integrate the banking reconstruction, AI/MCP incident and Knowledge/GitOps work, including demonstrations and operational guides, into the main release line.
  • Provide a versioned OpenShift client kit with immutable OCI images, Linux binaries, migrations, Helm chart, installer, test recipe, SBOM and provenance.
  • Gate publication on source tests, image and full dependency scans, Windows, Compose and real OpenShift qualification of the exact release artifacts.
  • Include PostgreSQL in image publication and MCP in native release archives.
  • Align Cargo, frontend and Helm versions. Keep prereleases out of stable/latest.

Safety fixes and limits

  • Application deletion is transactional; operation locks and retained Knowledge or reconstruction history return a conflict without partially deleting configuration.
  • Refresh vulnerable runtime bases and dependencies. Azure startup requires verified TLS enrollment and does not log enrollment tokens or fall back to an unauthenticated configuration.
  • Document maintenance interruptions and database recovery explicitly. The initial OpenShift profile is single-backend; it does not certify high availability, customer integrations or recovery of destructive operations.

OpenShift evaluation upgrade note

  • Backend HTTPS egress is now disabled by default. Before upgrading an existing OIDC/SAML/AI-enabled installation, configure networkPolicy.backendHttpsCIDRs for its approved destinations, or explicitly retain the former broad behavior with networkPolicy.allowAnyBackendHttps=true. Otherwise external authentication or AI calls can fail. The isolated password-login evaluation requires neither.
  • The client installer accepts a non-secret JSON configuration, verifies the live gateway connection and exports a public agent CA. External agent enrollment and immutable/offline image transfer procedures are included.

  • Retain two old Deployment revisions by default (global.revisionHistoryLimit), preventing repeated evaluation rollouts from accumulating the Kubernetes default of ten ReplicaSets per service.

Dependency security

  • Update JWT, TLS, randomness and frontend tooling dependencies; validate optional JWT not-before claims, including malformed values.
  • Scan complete workspace manifests in CI in addition to image OS packages and RustSec; retain development findings and reject production vulnerabilities.

Connection recovery

  • Late cleanup of an old gateway socket cannot remove its replacement connection or agent routes. Both socket tasks are stopped before cleanup.

Added

  • Phase 7: Production readiness improvements
  • Migration auto-run: Backend automatically runs sqlx migrations on startup
  • Graceful shutdown: SIGTERM/SIGINT handler with in-flight request draining
  • Configurable CORS: CORS_ORIGINS env var replaces CorsLayer::permissive(); restrictive by default in production
  • Structured JSON logging: LOG_FORMAT=json enables structured JSON logs for log aggregation (ELK, Loki, Datadog)
  • Prometheus metrics: /metrics endpoint with http_requests_total, http_request_duration_seconds, ws_connections_active, agents_connected, state_transitions_total, commands_executed_total, db_pool_connections
  • mTLS fingerprint forwarding: Gateway extracts and forwards agent certificate fingerprints to backend for identity binding
  • Auto-partition maintenance: Background task creates check_events partitions for current + next year; daily maintenance loop
  • React ErrorBoundary: Catches rendering errors in frontend subtrees with retry button
  • Frontend test infrastructure: Vitest + React Testing Library with unit tests for stores, hooks, and components
  • .env.example: Complete reference for all environment variables
  • CHANGELOG.md: This file
  • Phase 8: Production hardening
  • mTLS proxy fingerprint extraction: Gateway extracts X-Client-Cert-Fingerprint header from TLS-terminating proxies (nginx/envoy), overriding agent-reported fingerprint for tamper resistance
  • Security headers middleware: X-Frame-Options DENY, Content-Security-Policy (default-src 'self', frame-ancestors 'none'), HSTS (31536000s + includeSubDomains), X-Content-Type-Options nosniff, X-XSS-Protection, Referrer-Policy, Permissions-Policy
  • JWT HttpOnly cookie support: Multi-source auth (HttpOnly cookie > Bearer header > API key), SameSite=Strict, Secure in production; OIDC and SAML callbacks set cookies directly
  • Token revocation via PostgreSQL: Token fingerprint blacklist checked on every request
  • Kubernetes NetworkPolicies: Backend ingress from frontend+gateway only, frontend ingress from ingress controller, gateway ingress from any (external agents); egress restricted per component
  • Bounded channels with backpressure: Gateway uses mpsc::Sender (1024 agent, 4096 backend) with try_send() — drops messages on full channels to prevent OOM
  • Retransmission with deduplication: DeduplicationTracker with per-agent sequence ID tracking, high watermark advancement, gap detection, ring buffer of 1000 IDs per agent
  • Process kill on timeout: SIGTERM to process group, 5s grace period, then SIGKILL; returns exit_code -1 on timeout
  • Trivy vulnerability scanning in CI: CRITICAL+HIGH severity scan on all 4 Docker images, SBOM generation (CycloneDX format)
  • Structured error handling: ApiError enum (Database, NotFound, Forbidden, Unauthorized, Conflict, Validation, Internal, ServiceUnavailable) with PostgreSQL unique violation → 409, structured JSON error bodies, input validation helpers
  • Configurable database pool: DB_POOL_SIZE (default 20), DB_IDLE_TIMEOUT_SECS (default 600), DB_CONNECT_TIMEOUT_SECS (default 30), max_lifetime 30min, pool metrics reported to Prometheus every 10s
  • Prometheus metrics instrumentation: metrics_middleware records http_requests_total (method, status) and http_request_duration_seconds (method, path) with UUID path normalization; WebSocket and pool gauges updated every 10s
  • Graceful shutdown with configurable timeout: SHUTDOWN_TIMEOUT_SECS (default 30) prevents indefinite hangs on SIGTERM
  • Data retention policies: RETENTION_ACTION_LOG_DAYS and RETENTION_CHECK_EVENTS_DAYS (0 = unlimited), daily background task drops old check_events partitions
  • QoS message priority: MessagePriority enum (Low for heartbeats, Normal for configs, High for commands, Critical for register/approvals) on all WebSocket messages
  • SIGHUP config reload: Agent reloads config file and updates log level on SIGHUP signal
  • Pod anti-affinity rules: preferredDuringSchedulingIgnoredDuringExecution (weight 100, hostname topology) on backend, gateway, and frontend deployments
  • 60%+ frontend test coverage: 15+ new test files covering API client, stores, hooks, components, and pages (up from ~5%)
  • Phase 9: Sharing, API keys & documentation
  • Workspace validation for permissions: validate_workspace_access() checks target user/team has site access before granting permissions or consuming share links
  • User search/discovery with autocomplete: GET /users/search?q=&limit= endpoint (ILIKE on email + display_name, org-scoped) with UserPicker typeahead component
  • Share link consumption flow: POST /share-links/consume (validates token, expiry, max uses, workspace, grants permission), GET /share/:token preview (unauthenticated), accept/decline UI page
  • Combined permissions API: GET /apps/:app_id/permissions (users + teams joined with details), DELETE /apps/:app_id/permissions/:perm_id (tries user then team)
  • API key management UI: Full CRUD page — create key with copy-once warning, list keys (prefix, scopes, status, dates), revoke; linked from Settings
  • ShareModal improvements: UserPicker autocomplete replaces free-text input, share link revocation button, copy-to-clipboard feedback
  • Comprehensive documentation: docs/QUICKSTART.md (Docker Compose, local dev, auth setup), docs/USER_GUIDE.md (all features, operations, sharing, API keys, administration with screenshot placeholders), CLAUDE.md documentation maintenance instructions
  • CI screenshot auto-generation: Playwright captures all pages (1440x900, light theme), CI workflow builds stack, captures screenshots, embeds in docs, auto-commits
  • Phase 10: Production review fixes
  • Rebuild engine waits for command completion: Polls command_executions table, waits for each rebuild command to complete or timeout before proceeding; suspends on failure instead of blindly restarting
  • Switchover SYNC waits for integrity check results: Dispatches integrity checks AND polls for results; fails the SYNC phase if any check returns non-zero or times out
  • FSM database transactions: transition_component() wraps state read + validation + INSERT + UPDATE in a transaction with SELECT ... FOR UPDATE to prevent race conditions
  • cargo-audit enforcement in CI: Removed continue-on-error: true — vulnerability findings now fail the build
  • Action log archival instead of deletion: Retention task archives old entries to action_log_archive table instead of deleting, respecting append-only compliance (Critical Rule #2)
  • Webhook circuit breaker: After 5 consecutive failures, webhook is skipped for 5min cooldown; success resets circuit; lock-free global state via LazyLock + DashMap
  • PostgreSQL-backed rate limiting: HA_MODE=true uses PostgreSQL INCR+EXPIRE pattern (safe across replicas), falls back to in-memory DashMap otherwise
  • Diagnostic query optimization: Single query with ROW_NUMBER() OVER (PARTITION BY component_id, check_type) replaces O(3N) individual queries
  • Dead code cleanup: Removed unused variables in switchover config swap
  • Auto-init PKI at startup: auto_init_pki() generates CA for organizations without one at startup — zero-config mTLS
  • Database-backed authentication state: Revocation uses revoked_tokens; shared rate limiting uses application database counters.

[0.1.0] - 2026-02-22

Added

  • Initial release of AppControl v4
  • Common crate: FSM (8 states), protocol types, mTLS PKI utilities
  • Agent: Process execution with double-fork + setsid, offline buffer (sled), multi-gateway failover, native commands (disk/memory/cpu/process/tcp/http), resource limits (RLIMIT_CPU/AS/NOFILE/NPROC)
  • Gateway: Agent registry, message routing, backend auto-reconnect, agent rate limiting
  • Backend API: 75+ REST endpoints under /api/v1/
  • Applications CRUD with DAG-based start/stop/restart
  • Components CRUD with FSM state management
  • Teams, permissions (5 levels), share links
  • DR switchover (6-phase engine)
  • 3-level diagnostics + rebuild orchestration
  • DORA-compliant reports (availability, incidents, switchovers, audit, compliance, RTO)
  • Scheduler integration (orchestration API + CLI)
  • Variables, groups, links, command parameters
  • YAML map import (old AppControl format)
  • 4-eyes approval workflow
  • Break-glass emergency access
  • SAML 2.0 + OIDC + API key authentication
  • Workspace-site access control
  • Host-based agent resolution
  • CLI: appctl binary (start, stop, status, switchover, diagnose)
  • Frontend: React 18 + TypeScript + Tailwind + shadcn/ui + React Flow
  • Dashboard with weather cards and KPIs
  • Interactive DAG map with component nodes
  • Team management
  • Agent monitoring
  • DORA reports
  • YAML import
  • Onboarding wizard
  • Database: 14 PostgreSQL migrations, partitioned event tables, append-only audit trail
  • Docker: Multi-stage builds (backend, frontend, gateway, agent), 3 compose variants (dev, build, release)
  • Helm: Kubernetes chart with OpenShift compatibility
  • CI/CD: Build + test + lint + security scan, multi-platform release pipeline
  • Security: JWT RS256, rate limiting (3-tier), config version snapshots, heartbeat monitoring
  • Tests: 134 Rust unit tests, 183 E2E test functions