GridWeave is under construction and not released. This page says which milestones are built, which are partial and which have not started, and lists the gaps without softening them. It gives no dates, because the project has no schedule.
State of the product
Everything below is merged into the main branch of the product repository as of 2026-11-13. “Proven” means an automated test or a recorded run, and the notes say where each was run.
Multi-tenant data model with row-level security on every tenant table, local accounts with a second factor, OpenID Connect, LDAP and SAML 2.0 sign-in, roles, invitations, personal access tokens, hash-chained audit. LDAP and SAML were proved against OpenLDAP and Keycloak in containers, not Active Directory.
Enrolment, mutual TLS, certificate renewal, revocation, presence, reports, and server-signed commands that the agent verifies itself. The agent runs as a service. Proven on Ubuntu 24.04 ARM64; the macOS service was only dry-run; the Windows service was proven in a virtual machine that is now unreachable.
A relay with signed single-use grants and consent on the device. The real agent and server ran terminal and files sessions together on macOS only. The web tabs and the command-line client ran against fake agents and by hand in a headless browser.
Terminal session recording with integrity checks, a playback API and player, and guest links, on the server and in the web application. Terminal sessions only. Proven against fake agents, the real relay and a real S3 service; not with a real agent or a real browser, and the agent shows no notice yet.
Rollouts in rings with windows, concurrency and a health gate, and an agent that downloads, verifies and applies an update itself, with automatic rollback by the new build or a watchdog. Proven on Linux under systemd; macOS launchd and Windows are unverified.
Server and agent images, a split-role Compose stack, a Helm chart run on one single-node cluster (evaluation, not production), an OpenBao key store tested against a real OpenBao (not the default), and S3 object storage tested against SeaweedFS.
React and TypeScript in English and German, redesigned with Tailwind CSS and shadcn/ui and merged. An optional layout check opens every screen at three widths in both themes. Tests ran against stubbed answers, not a real browser against a real agent.
An optional role with a CIRA listener, inventory, power control and policy, built and tested against a simulator only. Nothing is validated on hardware.
A simulated-fleet generator and a gateway conformance kit. 10,000 simulated agents were held once on one shared Mac. No run has been made on a reference Linux host.
The product has run on one macOS machine and one Ubuntu 24.04 ARM64 virtual machine. Nothing has run in a customer environment. See the features page for the per-area detail.
Milestones
The delivery plan works in milestones. The labels are the ones the product's own plan uses. The web application was also redesigned and merged, and the binaries were renamed. Statuses are the product's, not a summary that rounds up.
| Milestone | Holds | Status | What is true |
|---|---|---|---|
| M0 Reference lab | Validation against a running system of the earlier product, on real agents | Not done | Outside the product repository. It gates only M6. Work that depends on it is built from the specification and marked unvalidated. |
| M1 Control plane | Server, schema, tenancy, accounts, OpenID Connect, sessions, second factor, roles, outbox, events, audit, web shell | Built | Done except the listed gaps (see below). Two role-based access requirements moved to M6. |
| M2 Agent plane | Gateway, mutual TLS, renewal, revocation, presence, reports, typed commands, the agent as a service on three operating systems, signed installers, conformance kit | Partial | Built, but the exit criteria are not met. Presence now renews on any frame and its lease follows the heartbeat. Enrolment and connection conformance pass at the gateway; the authorization and operations suites on three operating systems and the 10,000-agent run on a reference host were not run. Nothing is code-signed. Security review 1 is done and its findings are closed except the web half of one. |
| M3 Terminal and files | Relay, terminal, files, consent, access tokens, command-line shell and copy, desktop prototype | Partial | Built except the desktop prototype. Proven end to end with the real agent on macOS only; the Windows parts were never compiled or run. |
| M4 Desktop, recording, sharing | Remote desktop, recording, guest sharing, peer-to-peer upgrade | Partial | Terminal recording and guest sharing are built on the server and in the web application. Remote desktop, desktop recording and the peer-to-peer upgrade are not built, and the agent shows no recording or guest notice yet. |
| M5 Management completeness | Notifications, passkeys, password reset, invitations, agent releases and staged rollout, reports, bridge, 24 languages | Partial | Password reset, e-mail verification, invitations, stored notification subscriptions, agent releases (development keys) and staged rollout with agent self-update are built; self-update is proven on Linux under systemd only. Passkeys, self-registration, notification delivery, reports, the bridge role, enrolment links and 24 languages are not. |
| M6 Legacy adapter, importer | An optional compatibility adapter for a previous product, and an importer | Not built | Not started. It waits on M0. |
| M7 Parity, extensions | Extensions framework, LDAP and SAML, Helm chart | Partial | LDAP and SAML 2.0 sign-in, the Helm chart and S3 object storage are built. The extensions framework and the rest are not started. |
| M8 General-availability hardening | Scale, soak and chaos testing at 100,000 devices, disaster-recovery drill, external penetration test | Not built | Not started. |
| M9 Intel AMT | Optional Intel AMT module | Simulator only | The basics were brought forward and built against a simulator. Nothing is validated on hardware. Remote desktop and serial-over-LAN, activation policy, the web screens and the agent side are not built. |
Order, not dates
The plan sequences the milestones. It carries no calendar. The weeks in the original specification are indicative figures for a larger team and are not a schedule for this project.
In flight
Known gaps and risks
Do not use GridWeave to manage devices you depend on.
The product is not released and has had no external security review.
| Gap | Status | Detail |
|---|---|---|
| Code signing | Not done | Nothing is signed: no Authenticode or Developer ID certificate exists, installers are unsigned, the update-signing keys in the repository are development keys and the offline key ceremony has not been done. There is no install page or enrolment link. |
| Continuous integration | Disabled | Workflows are prepared and switched off until a billing problem is fixed. No hosted run has ever happened. The local check set is the gate. |
| Windows | Unverified | The Windows 11 virtual machine became unreachable on 2026-11-13. The Windows terminal and consent prompt were never compiled or run, the Windows installer script was never run, nothing swaps a binary on Windows for self-update, and several tests did not complete. |
| macOS service | Unverified | The launchd service and the self-update watchdog job on macOS were never run, only the files they write. |
| Web enrolment of a second factor | Open | The server requires the current password when a TOTP factor is added. In the current web source the enrolment dialog does not send it, so enrolling from the screen is refused until the web side is updated. |
| Intel AMT hardware | Simulator only | Every behaviour that depends on firmware is unvalidated. AMT remote desktop and serial-over-LAN, activation and the AMT web screens are not built. |
| Remote desktop | Not built | Neither the prototype nor the full feature exists. |
| Notification delivery | Not built | Subscriptions are stored. Nothing is delivered. |
| Legacy adapter | Not built | The optional compatibility adapter for a previous product and its importer are not started. |
| Scale run on a reference host | Not done | 10,000 simulated agents were held once on a shared Mac. No run on a reference Linux host, and the 50,000-agent target was not run. The heartbeat and presence-lease mismatch is fixed (feature 042), but offline detection takes up to 180 seconds at a 60 second heartbeat, and the fix was not re-measured at scale. |
| External penetration test | Not done | Security review 1 was done inside the project and found no critical finding. An external test is part of M8 and has not happened. |
| Binary names | Done | The binaries were renamed to gridweaved, gridweave, gridweave-admin and gridweave-agent (feature 043). The Windows service names were only type-checked and the macOS launchd names were checked with plutil, not run. The NS_* variables, ns.* protocol packages and ns_ database roles kept their names. |
| Other partial items | Partial | The production key store is not the default. The agent keeps its key in a protected file, not in a hardware or platform store. The web Files tab has no folder browsing. Command-line login is by token only. There is no helper application for consent prompts on Linux and macOS desktops, so the default is to refuse when nobody can be asked. Passkeys, self-registration and reports are not built. |
Before a first release
These are the items the product’s own status documents name as needing the owner or outside resources. Having them does not make a release a promise, and there is no date.
Authenticode and Developer ID certificates, so installers can be signed.
The offline update-key ceremony, replacing the development keys pinned in the repository.
A Linux host for the 10,000 to 50,000-agent runs, and the conformance suites on three operating systems.
An independent penetration test. The internal review is not a substitute.
Intel vPro hardware to validate the AMT module beyond the simulator.
A Windows test machine to verify the Windows parts, and the CI billing fixed so hosted runs can be enabled.