Release history

Adds a CSAF/VDP disclosure sensor: probes each vendor's standard CSAF discovery location alongside the existing security.txt check, both counting as evidence of a NIS2 Art. 21 coordinated vulnerability-disclosure process, and opens a Risk when a critical vendor has neither. Adds SBOM vulnerability scanning - periodically re-checks each vendor's published SBOM against OSV.dev and opens a Risk Register entry for every finding, notifying the system owner by email and/or Slack/Teams once a finding reaches an admin-configured minimum severity from the new Settings > Notifications tab. When a vendor also publishes a CSAF distribution, findings are cross-checked against the vendor's own VEX advisories and silently skipped when the vendor has classified that exact CVE/purl pair as known_not_affected - matched only by an exact purl hit against the vendor's own product tree, never guessed from a name or version. Adds role-based access control: logging in no longer requires admin-group membership, any directory account can sign in, but a non-admin only sees and manages the systems it owns, while the admin group keeps full access to everything including Settings and the other admin-only pages, which are now hidden from non-admin navigation; a new system's owner is locked to the creating user, and an owner can never reassign ownership. A background security review of this access-control work caught a real IDOR where several create endpoints let a non-admin's client-supplied ID silently overwrite an existing System, Risk, or Contact record - fixed by always issuing a fresh server-generated ID for non-admin sessions.

Full changelog · ab8445022b70a0b4ef2627f606fc810c108908372f0cd026bbf60b17442eb6e5

Adds end-to-end-encrypted settings sharing with Allod SWG (Pull from SWG for IDP/OIDC and S3 storage config, sealed with nacl/box between each instance's own X25519 identity, TOFU-paired and failing closed on identity mismatch) and, alongside it, VRM's first real S3 admin surface with hot-swappable document storage so vendor e-signing activates without a restart. Fixed the SWG shadow-IT sync pipeline end to end: it had been failing on every cycle in production because the loopback SWG API is TLS-wrapped with a self-signed cert the sync client had no way to trust; sync status is now tracked and surfaced persistently in the UI instead of only reaching a log line, and UI-saved SWG settings are read back at startup instead of reverting to the static config file on every restart. Systems that resume traffic while offboarding, decommissioned, inactive, or rejected are now handled correctly - auto-reverted or notified rather than silently ignored - and rejected systems can be reactivated from the Triage page. Added an LDAP/Active Directory login backend alongside the existing OIDC option, with a Test Login panel that verifies real credentials against the saved connection (bind, resolved groups, and admin-group membership) without creating a session, and merged the standalone Auth tab into the directory/SCIM tab since they're the same connection. The top navbar's Triage, Classification, Cycle Types, and Contract Templates now live under one Admin dropdown instead of eight flat items - which surfaced a real bug where the dropdown never actually opened: bootstrap.bundle.min.js and tabler.min.js were both loading a full copy of Bootstrap's JS, so every click opened the menu and an unaware second listener immediately closed it again. Also fixes deleting a system leaving its scheduled reviews, vendor probe results, and RoPA data permanently orphaned - some of those tables never had a foreign key back to the system at all, and SQLite doesn't enforce the ones that do by default.

Full changelog · 6411d724978232f092ece665886c37d0cf81013353d137382f157b08ff034354

Renamed from Allod DAM to Allod VRM (Vendor Risk Management), reflecting where the product has grown since v0.1.0. Sovereignty scoring saw several correctness fixes: a weighted-sum renormalization bug that inflated scores when not every objective had been assessed, a dead control, redundant SOV-4/6 checks, and closed content gaps in the CSF §4 objectives (SOV-1/3/5/8) - plus manual Vendor HQ and Ultimate Owner country fields and an optional self-hosted-deployment signal that feeds SOV-3/SOV-4 suggestions without ever setting a level on its own. Added vendor e-signing for contracts: an admin uploads a document unmodified, emails a one-time link, and the vendor's approval is captured as a separate Ed25519-signed receipt (document hash, signer, IP, timestamp) rather than altering the original file. Added Contract Templates: reusable HTML or Markdown templates stored in S3-compatible object storage with bucket versioning as the version-history mechanism, an explicit publish step so in-progress edits never affect what a contract is generated from, and a sanitized (goldmark + bluemonday) preview for Markdown templates rendered in a sandboxed frame. Fixed a stored-XSS vulnerability from unescaped single quotes and added missing foreign-key validation.

Full changelog · eefcc2eef1bbe299c64bdbaa3ce7b75ca72f1b201efb2427cdc31a2ec01bd1f7

First tracked release of Allod DAM (Digital Asset Manager) - a self-hosted vendor governance and supply-chain visibility platform. Tracks a system inventory across owner, category, criticality, contract/license dates, and GDPR RoPA (processing role, DPA status, sub-processors, third-country transfers). Vendor security posture is continuously assessed via around 23 automated probes (TLS, DNS/mail authentication, security headers, SBOM presence, company registry data) and scored against the CSF SEAL v1.2.1 sovereignty framework across 8 weighted objectives - strategic, legal, data & AI, operational, supply chain, technology, security, and environmental. Imports and tracks published SBOMs (SPDX/CycloneDX) and vendor OSCAL documents (component-definition and system-security-plan). Discovers shadow IT and uncatalogued internal tools directly from Allod SWG's observed traffic. Named, admin-configured review-cycle templates - interval, contract-renewal, license-renewal, or manual triggers - drive a scheduled reassessment workflow, and SCIM v2 keeps user and group state synced from the customer's own identity provider. This release adds a formal Risk Register: risks are automatically opened and kept current from sovereignty scores, missing DPAs, absent SBOMs, and approaching contract or license expiries, each with likelihood/impact scoring, a mitigate/accept/transfer/avoid treatment workflow, time-boxed risk acceptance, and residual-risk tracking - without ever overriding a decision a human has already made on an entry.

Full changelog · 966f51d18e7764a0413b626b544fdf114dce2acc4044947ef3bcc82840ffa2db

The controller can now run as multiple replicas for high availability: with a shared Postgres event store, admin sessions, API keys, both CAs, DLP encryption keys, and all other previously in-memory or file-local state are backed by Postgres, one replica is automatically elected leader for background work, and a leader-aware health check lets a load balancer route traffic only to the active replica - failover has been verified with a real test that kills the leader mid-operation and confirms the standby takes over within seconds. Every downloaded agent, allodctl, and server release artifact is now cryptographically signed and verified before use, rejecting fail-closed on a missing or invalid signature. Security advisories for any reachable, fixed vulnerability are now drafted and published automatically at release time instead of by hand, alongside a new published vulnerability-disclosure policy and support-period statement, and every container image is now scanned for known CVEs before publishing rather than only inventoried. The proxy and the agent-facing management API are also now hardened against connection floods: traffic that never presents valid credentials is identified early and dropped, before it can consume the same processing resources as legitimate agent traffic, including floods spread across many source addresses rather than just one - this protection is opt-in and does not change existing behavior unless explicitly enabled. Fleet admins now get a proactive notification (via whichever alerting channel is already configured) when a new agent version becomes available, and can postpone an automatic fleet-wide rollout for a chosen number of hours instead of only being able to disable automatic updates entirely. Also fixes two real incidents: a controller could end up serving a stale, already-superseded signing CA instead of the one it actually distributes to devices, and migrating an existing installation's CA into Postgres for the first time could silently mint a brand new, unrelated CA instead of preserving the one every OS, browser, and agent already trusts - both causing hard TLS failures through the proxy. Firewall and DLP rule changes made via the admin UI now persist across a container recreation instead of silently reverting; several further dependency CVEs closed.

Full changelog · 44c7a3805d88d2cf9b01850da3ce38ea4a683b3000af4fa1855c2ee3417a5116

The Branding settings page (formerly Block Page) now includes a live, multi-page preview so you can see exactly how your logo and colors will look across the sign-in form, step-up access pages, and self-service portal before saving. Agents now detect OS-level sleep/hibernation directly and reconnect the moment the device wakes, instead of waiting out the normal retry backoff, with a WebSocket keepalive as a fallback for any network loss that mechanism misses. Self-hosted git servers reached over ssh:// are now correctly recognized as company-owned destinations - this fixes a false CASB/DLP flag and a garbled URL on the DLP dashboard, and surfaces these internal git hosts in VRM triage and Corporate IT for the first time. Server, proxy, and connector are now also available as container images, each shipping with an attached CycloneDX SBOM for supply-chain verification.

Full changelog · bfd132755dbdf9f823062176fe504d0a15c08e4bad0fe6bbe512f9581f7c0dcc

The LDAP/Active Directory sign-in form, the identity-verification and step-up access result pages, and the self-service access-request portal now show your organization's own branding - logo, colors, name - the same settings already used on the block page, instead of generic AllodSWG chrome. Also fixes a false positive in the agent's supply-chain protection that could kill an in-progress go build or go test (or any tool it spawns, including headless-browser tests) on a developer's machine, and a bug where blocked supply-chain events weren't showing up in the admin events log.

Full changelog · 34b9db81d97a68fbd98cd62c7dfa16eb99dcc8e03ece1391672f2fd08cceee66

The IDS/IPS engine now supports Suricata's base64_decode/base64_data keyword pair - rules that match content against base64-decoded payloads (a common technique in real feeds like URLhaus, for obfuscated PowerShell/webshell/dropper content hidden in form fields or headers) previously parsed but were silently never evaluated; they now match. The decoder is lenient like Suricata's own (keeps the valid prefix, stops at the first invalid byte, rather than discarding the whole decode on the first bad character).

Full changelog · a5e327abeb63e88cc8efb71f101959be2b7f147f3017bb5085fa79c6183ca4ac

QUIC/HTTP-3 inspection now works on Windows, not just Linux: the agent redirects outbound UDP/443 to a local relay via WinDivert (the Windows equivalent of Linux's nftables DNAT), which terminates QUIC with a per-hostname MITM certificate and forwards to the proxy's QUIC inspector with the same X-MyrSWG-* identity headers HTTP/2 traffic already gets - verified end-to-end on a real Windows 10 machine, including a real headless Edge browser actually negotiating HTTP/3 over the locally-installed CA. Security fix: quic_mode - which controls whether QUIC traffic is inspected, downgraded to inspectable TCP, or left alone - defaulted to fully uninspected ("bypass") on any fresh install that never explicitly touched the setting; now defaults to "downgrade", so an unconfigured or older agent never silently lets QUIC traffic through unread. Chrome/Chromium on Linux now also gets its CA trust written into its NSS certificate database (the store it actually reads on Linux, not the OS trust store) as a supplement to the existing managed-policy approach, which only newer Chrome versions honor. Screenshot and print-job OCR (used for DLP content matching against on-screen or printed text) now actually works: replaced a tesseract-CLI dependency that was never packaged or installed anywhere - meaning OCR had silently been a no-op on real fleets - with go-ocr, a new cgo-free library built for this that runs PP-OCRv5 via ONNX Runtime, published as an open-source Allod Solutions project.

Full changelog · 994d369c37e803a4c2c6282ddb17627cd03ba07dd9fb742f5b6090206653189f

Container-aware egress attribution now works on Windows and macOS (both via Docker Desktop), not just Linux - firewall rules can scope by container/pod label and DLP events show container name/image/labels on all three platforms, though Windows/macOS support ships marked experimental: correlation can miss when a container's traffic is routed through the agent's own system-proxy path instead of being transparently intercepted, so pair it with a process- or destination-based rule as a backstop there. Getting this working surfaced and fixed several real Windows bugs found only by running the mechanism against a live container in a golden-image test VM: a PID-namespace mismatch in Docker's own API response, wsl.exe silently stripping shell variables from generated scripts, a missing auth header on WinDivert asset downloads, a WinDivert handle-priority collision, an unsupported processId filter field, and a filter-negation syntax WinDivert rejects - the net effect is that Windows HTTP/HTTPS transparent inspection now actually opens for the first time. New: browser-extension request detection flags traffic originating from a Chrome/Edge/Brave/Opera or Firefox extension's own background script (via the Origin header, since TLS is already terminated) - useful against a malicious or compromised extension exfiltrating page content from other open tabs, since that traffic otherwise looks identical to ordinary browsing. Supply-chain protection's installer-session tracking is now liveness-based instead of a fixed 5-minute timer, closing a bypass where a malicious postinstall script could sleep past the old window and read credentials undetected; it also now detects a tracked process detaching to persist independently (daemonizing), with existing process-allow-rules already covering legitimate watch-mode tools that do this on purpose. Also: the device-identity sign-in page now returns you to the site you were trying to reach after verifying, instead of a dead-end 'you can close this window' page; several dashboard card layout fixes.

Full changelog · b3f83e62703238680154f52262d4ba7ceb678b5251826a277789855af562f278

LDAP joins external OIDC as a full login backend: the Auth tab can now point admin/approver login, step-up SSO/JIT, device-identity binding, and the allodctl CLI at a directory bind instead of an external IdP, with a Test Login button that verifies real credentials and group authorization before you commit to it. Directory Sync auto-discovers LDAP base DNs and detects the schema flavor (posixAccount/AD/etc.) with a one-click Detect Schema button, fixing a group-membership foreign-key bug along the way, and the Pull-from-VRM settings action now carries LDAP connection fields and reports the peer's actual auth state honestly instead of guessing. Settings can now be shared end-to-end encrypted directly between SWG and VRM. Email authenticity scoring folds in typosquat-link detection and tunes down the HTML/plain-text mismatch weight so it needs corroboration, reducing false positives. The dashboard merges JA4/JA3 into a single Fingerprints card and folds IDS/IPS into Threat Intel. New: a device that fails identity verification now lands on a real sign-in page on its very next request instead of getting a silently dropped connection, and the Linux desktop notification for identity verification is finally clickable. The LDAP sign-in page itself has also been restyled to match the rest of the admin UI instead of a generic unstyled form. Also hardens the Linux identity-verification notification against argument injection via a crafted filename or email subject in its title/body.

Full changelog · 841eeee5d8e3e9dba99cd1c82fb05d885cf5f04ad48c18256550ce6cea505cf4

Suricata rules gain flowbits/flowint support for multi-stage correlation within a single connection - a low-severity signal can mark a flow without alerting on its own, and a later rule can escalate only once that mark is present. A third rule source joins custom rules and the abuse.ch URLhaus feed: Allod's own vendor-curated ruleset, delivered and refreshed through the same license check-in mechanism. The Firewall page's custom-rules card now adds and edits rules one at a time with a small input, mirroring the firewall rules table's UX, instead of a paste/upload-only textarea; bulk replace-all import remains available but API-only. Fixes a bug where disabling IDS/IPS after enabling it never actually stuck in the UI. The dashboard's blocked-countries map no longer lights up a country for a device blocked there by an unrelated rule that had nothing to do with its location. New: detection and active blocking of supply-chain credential theft - a compromised npm/pip/yarn/etc. package's postinstall script spawning a secret scanner (TruffleHog and similar) to read .env files, SSH keys, or cloud credentials is now caught and killed before it can exfiltrate anything, with an admin-configurable exemption for legitimate build tools that genuinely need one of those paths.

Full changelog · 0602ac3e55b685e72077adc5a5364e6b4ec430099202a77442148efa87dd4ff2

Adds a Go-native Suricata-compatible rule engine for the SWG's IDS/IPS layer, off by default: admins can author and import their own .rules-syntax content/header signatures directly on the Firewall page, alongside a curated abuse.ch URLhaus feed of pre-vetted rules that ship pinned to block regardless of the feed's own declared action. Header/SNI/IP/port rules are evaluated at CONNECT time in the same detector chain as existing JA4/JA3/DGA/typosquat matching; content, pcre, and http.*-buffer rules are evaluated against the request/response body and merged with DLP's own verdict. A pcre: pattern using backreferences or lookaround - constructs Go's native regex engine can't represent - now falls back to a second, backtracking-capable engine instead of being silently skipped. The device/threat map on the admin dashboard can now be embedded in an external dashboard or wallboard via a separate, token-authenticated public route - data is always aggregated to country granularity, with no device names, usernames, or exact coordinates ever exposed - and includes a blocked-country red-zone overlay. Inbound email inspection now recognizes a native "Report spam"/"Report phishing" action taken directly in Gmail's webmail UI or via IMAP, not just a forwarded copy.

Full changelog · 1e2a640d4304e034692bf155d11ae9cc3c6ff7d7c54d8cd2dd183bed20cb1246

Captive-portal detection on the endpoint agent is dramatically faster - down from roughly 90-100 seconds to consistently under 20, confirmed against real hotel wifi. The agent now probes for a portal immediately after a network change instead of waiting on the ordinary heartbeat cadence, trusts a TLS certificate hostname mismatch as direct evidence of a portal, and exits bypass mode the instant connectivity is confirmed restored rather than leaving interception paused for up to another heartbeat interval. Fixed a bug where TLS interception - along with SMTP, mail-retrieval inspection, and ZTNA access routes, which share the same underlying ruleset - could silently stay disabled after a captive-portal bypass cycle even though the agent logged a normal configuration-complete message. Agent DNS resolution now falls back to public resolvers when the network-assigned one is unreliable, and enrollment retries start faster after a network change. The email-authenticity score behind a flagged inbound email is now persisted and shown in full in the admin UI, broken down signal by signal instead of just a total.

Full changelog · abda7ae16137ac008964a726b16ee8ea00bbe764c16189daaf68500e5a2c7125

The 1.0.0 general-availability release. The operator command-line tool is renamed from allodswg-ctl to allodctl and reworked around what is actually on the machine: the menu offers to install a server or an agent where none is found, and full status / upgrade / configure / uninstall management where one is. Both the allodswg-server and the endpoint agent now report their running version alongside the latest published version, so an available upgrade is visible at a glance, and a guided `allodctl server install` provisions the server monolith from the licensed distribution channel - now kept correctly running under systemd, where the service previously daemonized out from under its own unit. Inbound email phishing detection gains a signal for malformed header syntax and now recovers a Subject header that has whitespace before its colon. The compliance view's framework-coverage page is synced against the OSCAL control catalog. Internal DAM references are renamed to Allod VRM. GeoDNS and a dead agent-tunnel endpoint are removed, with proxy selection now documented as latency-based.

Full changelog · 80b7a7c11d2da8342f92b90dd4968e8323208dea06b89c6ad89f534141aad85f

Firewall and DLP rules scoped to a specific process now also match when that process appears anywhere in the connecting process's ancestry - closing a gap where a rule scoped to an application (e.g. an RDP or Citrix client) wouldn't catch a helper process it spawned that made its own network connection independently. The event log now shows the full process chain when a rule fires this way, so it's clear why. Strengthened inbound email phishing detection: a message that declares a non-standard content encoding on its HTML body - a technique seen hiding phishing links inside base64 content that scanners would otherwise skip - is now both decoded for inspection and flagged as suspicious in its own right. Added detection for GitHub and Slack API tokens to the DLP secret-pattern library, alongside the existing AWS and private-key detectors.

Full changelog · af26a4d8f80844627e211a34c0e6077e461323cb56c617faf6b9c885363a1e57

Added Google Drive support to the existing OneDrive/SharePoint sync-provenance and lineage-tagging feature: a file synced by Google Drive for Desktop's Shared Drives is now attributed back to its Shared Drive name the same way OneDrive files are attributed to a SharePoint site, including the ability to tag a specific Shared Drive with a standing DLP sensitivity rule. Strengthened inbound email phishing detection with three additions to the authenticity scoring engine: a fix so newly-registered-domain detection correctly evaluates the sender's actual registered domain rather than a subdomain of it; a new signal for links that serve a full HTML page directly from public cloud storage (Google Cloud Storage, S3, Azure Blob, and similar), a known technique for evading domain-reputation checks; and a new signal for messages whose plain-text part is trivial boilerplate while the real content only exists in the HTML part. Added a warning when a block-action DLP rule is scoped by a field that only resolves asynchronously after a transfer completes, since such a rule can silently never block. Added JA3 as an independently selectable scope condition on firewall and DLP rules, alongside JA4/Mercury/HASSH, and fixed the admin UI so the existing source-IP/CIDR scope condition is actually visible and editable on both the firewall and DLP rule pages.

Full changelog · 4153be53bd7caf75f30f6aabfa10b634530b32be8eedfa8964b1a34dd213bc19

Added SourceIP as a new scope condition on firewall and DLP rules - restrict any rule (domain, category, IP, region, ZTNA access, or content-inspection rule) to traffic from a specific source IP or CIDR range, alongside the existing device, user, and group scoping. Also improved ZTNA device-to-device access: a single-node deployment can now route directly to another enrolled device's exposed local port via its virtual IP, without requiring a separate connector daemon.

Full changelog · 59688fc67dd283022c56ede39a630650c5b9cd68ad912bc80c09f956e26c07ec

Added LDAP/Active Directory as a login option - no external OIDC or SAML identity provider required. Applies uniformly to admin console login, step-up SSO, just-in-time access approval, device identity verification, and the allodswg-ctl CLI. Reuses your existing Directory Sync connection settings, so there's nothing new to configure beyond selecting LDAP as the login backend under Settings. Also includes correct handling of Active Directory's default 1,000-account search limit for larger directories.

Full changelog · e60be37fd4cfd407f2eda5a73ef64d144f73dc762919c0d7e6f4476302c000b9