SentinelGo v3.5.0
Roadmap Native macOS MDM is next →

Every endpoint.
Seen, secured,
and kept that way.

SentinelGo is a single, dependency-free agent for Windows, macOS and Linux. It powers a full endpoint-management platform: live inventory, security posture and CVEs, compliance, policies, software deployment and remote actions, from one console.

Download v3.5.0 Explore the roadmap
Apache-2.0 agent No cgo, no runtime Signed self-update
24
agent releases since June 2026
3OS
Windows, macOS, Linux · 5 binaries
31
parsers fuzzed every night
0
runtime dependencies (no cgo)
8
native package managers
4GiB
verified installer uploads
Why SentinelGo

The agent you'd design if you had to trust it with root.

Most endpoint tools ship a heavy stack: a kernel module here, a scripting runtime there, a different installer per OS. SentinelGo goes the other way: one small, auditable, static binary that does its job and gets out of the way.

One static binary per platform

Every build is CGO_ENABLED=0, so there are no shared libraries, no interpreter and no kernel driver. A CI gate rejects any change that adds cgo. All targets cross-compile from one machine.

windows-amd64.exedarwin-arm64darwin-amd64linux-amd64linux-arm64

Updates itself, safely

The agent finds its next release through the backend, then checks the SHA-256 and Ed25519 signature. It replaces itself atomically and rolls back if the swap fails. It refuses unsigned updates.

Ed25519signed releases

A real OS service

Runs as a Windows Service, systemd unit or launchd daemon. It survives reboots, restarts on failure and shuts down gracefully.

Nothing gets dropped

Audit logs are checkpointed and queued on disk, then delivered at least once. Payloads survive outages and reboots, and retries are de-duplicated.

Private by default

No credential material is collected. Credential-like and PII patterns are redacted from task output before upload, and the agent only talks HTTPS.

The platform

The agent is the hands and eyes.
The console is the brain.

Most of the work happens server-side. The agent reports facts and carries out signed actions; the platform turns those facts into risk, compliance, alerts and automation.

Know every device, down to the partition.

A live, structured picture of the fleet that goes far beyond "is it online".

  • CPU, GPU, RAM modules, disks and partitions, displays, audio, printers, peripherals
  • Software for every user account: native, Store, Homebrew, APT/RPM/Snap/Flatpak, plus browser extensions
  • OS services (systemd, Windows Services, launchd), local accounts, network adapters and Wi-Fi
  • Device groups, tags, and devices mapped to the employees who own them

See risk before it becomes an incident.

Posture and vulnerability data, cross-checked against public sources.

  • Encryption per partition: BitLocker, FileVault, LUKS
  • Firewall, antivirus/EDR presence, Secure Boot, SIP, SELinux/AppArmor, TPM
  • Installed software matched to CVEs from the National Vulnerability Database
  • OS end-of-life tracking and patch status, with fleet-wide risk views

Compliance you define, evidence you can show.

Checks are configuration, not code: categories, severities, evidence rules and version operators, managed in the console.

  • Per-device pass/fail with the evidence behind it
  • Fleet percentages per check, ready for an auditor
  • New checks roll out without an agent release

Automate the boring, recurring work.

A visual policy editor and a drag-and-drop workflow builder for joiner/leaver and custom IT flows.

  • Policies assigned to device groups, run on a schedule or a recurrence
  • Workflows triggered manually, on a schedule, or by an alert
  • Steps to run commands, enforce policies, branch on conditions and wait
  • Full execution history for every run

Act on one device or a thousand.

PowerShell, Bash and shell scripts, reusable templates with variables, and built-in native tasks.

  • Target a device or a whole group
  • Describe a script in plain English and get a draft from the AI assistant
  • Native tasks such as reboot and firewall on/off, with per-task timeouts
  • Structured results: exit code, output, reboot required

Ship software the native way.

A curated catalog across eight package managers, plus your own installer files.

  • winget, Chocolatey, Homebrew, APT, DNF, YUM, Snap and Flatpak
  • Upload installers up to 4 GiB with resumable uploads; hashed on the server, verified on the device
  • Install from the catalog, on a single device, or through a policy
  • Uninstall, and keep the software inventory reconciled

From alert to fixed, in one click.

Rules built from device state, hardware, software and services, with a gallery of enterprise templates.

  • A rule wizard plus AI-assisted rule writing
  • Severity, acknowledge/resolve and recurrence tracking
  • Remediation commands linked to rules, runnable straight from the resolve step
  • Telemetry export to Wazuh SIEM over TLS

Least privilege, with a paper trail.

Enterprise access control for the people who manage the fleet.

  • Custom roles with a permission matrix grouped by resource
  • Scoped assignments: organisation, department, device group, site, device
  • Just-in-time roles with an approval flow
  • Full admin activity trail with diffs, saved filters and CSV export; Azure SSO
How it works

Facts flow up. Signed actions flow down.

The agent authenticates, reports, streams and keeps itself current, with no hand-holding after install.

  1. EndpointSentinelGo agent · Win · macOS · Linux
  2. heartbeat · inventory · logs ↓   ↑ signed tasks · updates
  3. Backendauth · queue · storage · release distribution
  4. risk · compliance · alerts ↓   ↑ policies · commands
  5. Consoleadmins · helpdesk · employee portal

Authenticate

Each agent signs in with its own identity and receives short-lived tokens, renewed in the background. A rejected session recovers once, automatically.

Report

A heartbeat every few minutes (default 5) plus periodic full hardware, software and posture inventory, with fast first sync after enrollment.

Act

The agent polls for tasks, fetches and verifies any installer files, runs the task with timeouts, and returns structured results.

Stay current

It checks for signed releases for its platform and architecture, verifies them, swaps itself atomically and restarts cleanly.

Supply-chain trust

Verifiable from commit to device.

An agent that runs with elevated privileges on every machine has to earn trust. Every release follows the same chain, and every link can be checked independently.

Source

Every change scanned and its untrusted-input parsers fuzzed.

CodeQLSemgrepgovulncheck31 fuzzers

Build

Static, cgo-free and reproducible: rebuild it yourself and compare.

CGO_ENABLED=0-race in CI

Sign

Binaries and installers signed, with published checksums.

Ed25519SHA256SUMS

Attest

Build provenance and a bill of materials for every binary.

provenanceCycloneDX SBOMlicence gate

Deliver

Served from the platform's own release storage in signed bundles.

per-arch bundlesHTTPS only

Verify

The agent checks hash and signature, swaps atomically and rolls back on failure.

refuses unsignedatomic swap
OpenSSF Scorecard SonarCloud quality gate License: Apache 2.0 Security assurance case published
Runs everywhere

Native on every OS your fleet runs.

Cross-compiled from a single host into static binaries. No per-platform build farm, no per-platform runtime.

Windows

Windows Service
Arch
amd64
Encryption
BitLocker, with central key escrow
Logs
Windows Event Log
Packages
winget, Chocolatey, MSI

macOS

launchd daemon
Arch
arm64 · amd64
Encryption
FileVault
Logs
Unified log
Packages
Homebrew, PKG

Linux

systemd unit
Arch
amd64 · arm64
Encryption
LUKS
Logs
journald, auth/syslog
Packages
APT, DNF, YUM, Snap, Flatpak
Public roadmap

From first commit to native MDM.

Everything we've shipped since May 2026 across the agent and the platform, where we are right now, and what comes next, including what we've deliberately decided not to build.

  1. 2026 · Delivered
  2. Platform

    Platform foundation

    The console is born: devices, users, remote commands, an app catalog, and vulnerability scanning on day one.

    What shipped
    • Admin console: dashboard, device list and detail, users, settings, device logs
    • Agent registration and per-agent installer issuing; employee directory sync and bulk import
    • Remote commands with AI-assisted script writing; app catalog; employee "My Devices" portal
    • Device groups, policies and alert rules
    • NVD CVE scanning, OS end-of-life tracking, agent releases page, integrations hub
  3. Agentv2.1.17 → v2.2.1

    One agent, three operating systems

    A cross-platform agent running as a native service everywhere, with rich inventory and audit-log collection on all three OSes.

    What shipped
    • Windows Service, systemd and launchd service modes
    • Extended network, system and timezone inventory
    • Reworked audit-log collectors: Windows Event Log, journald, macOS unified log
  4. Agentv3.0.0

    Signed self-update

    Every release is signed with Ed25519. The agent verifies checksum and signature before installing, and refuses anything unsigned.

  5. Agentv3.0.1 → v3.2.6

    Remote tasks, one-click installers, posture

    The agent learns to act: a native task framework, remote reboot and firewall control, double-click installers for every desktop OS, and cross-platform security-posture assessment.

    What shipped
    • Native task framework with per-task timeouts and a watchdog
    • USB mass-storage detection, peripheral and app-usage detection
    • Updates served from the platform's own release storage
    • Session recovery, emergency logging, de-duplicated delivery with retries
    • OS-services inventory, security posture and patch compliance on all three OSes
    • Diagnostics dump, redaction of sensitive values, software for every user account
  6. Platform

    Workflows and self-hosting

    A drag-and-drop workflow builder for joiner/leaver processes, and a Docker deployment with automatic migrations so the platform can run anywhere.

  7. Platform

    Enterprise access control

    Custom roles, scoped and just-in-time assignments with approvals, and a full admin activity trail.

    What shipped
    • Role-based access with a permission matrix, scoped assignments and JIT roles
    • Admin activity log with timeline, diffs, saved filters and CSV export
    • Fully configurable compliance checks; alert-rule wizard with a template gallery
    • Default "Standard" device group; generic workflows; first backup & restore
    • Agent v3.2.7: browser extensions reported with stable IDs and names
  8. Platform

    Endpoint Management

    The console grows up: a modular dashboard, BitLocker recovery-key vault, policy engine v2 and SIEM export.

    What shipped
    • Widget dashboard with OS breakdown, security overview and risk
    • BitLocker enablement with recovery keys in an encrypted vault
    • Policy engine v2: recurrence and app installs
    • Wazuh SIEM export over TLS; scheduled backups
    • Faster first sync after enrollment; more accurate patch alerts
  9. Agentv3.2.8

    Hardening

    HTTPS-only connections, restrictive permissions on the installation, system tools invoked by absolute path, and a code-quality sweep across ~30 files.

  10. Platform

    Production-ready deployment

    Software-inventory reconciliation, a first-admin setup wizard, a production container image and a dedicated installer project.

  11. Agentv3.3.0 · v3.4.0

    Supply-chain security and continuous fuzzing

    Provenance attestations and an SBOM for every binary, pinned CI, 31 nightly fuzz targets, and one in-house backend client.

    What shipped
    • OpenSSF Scorecard, SHA-pinned Actions, least-privilege CI tokens, licence allow-list
    • Build-provenance attestations and CycloneDX SBOMs per binary
    • 31 Go fuzz targets over every untrusted-input parser, run nightly
    • All backend access through one typed client with reliable session renewal and retries
    • 12 bugs found and fixed while raising test coverage; project website and governance
  12. AgentPlatformv3.5.0

    Installer files for every OS

    Upload installers up to 4 GiB, and have them hashed on the server, verified on the device, and installed from the catalog, a device or a policy. Built jointly against one written contract.

    What shipped
    • Resumable uploads, server-side hashing, install-command templates per installer type
    • Agent downloads and verifies artifacts before the script runs
    • Capability negotiation and structured results (exit code, reboot required)
    • Reproducible builds, race detection in CI, published security assurance case
  13. Platform

    Remediation, backups and sign-in hardening

    One-click remediation from alerts, rebuilt backup & restore (S3, SFTP, SMB), signed installer bundles and hardened sign-in.

  14. You are here · October 2026 · v3.5.0
  15. Now · Q4 2026
  16. PlannedAgent

    Security hardening

    Every remote-action path integrity-protected before the platform's powers grow further.

    Scope
    • Agent credentials in the OS credential store (Windows, macOS Keychain), without cgo
    • Signed task execution: database access alone can't produce a runnable task
    • Installation self-check at startup, reported fleet-wide
    Done when: unsigned or tampered tasks are rejected and reported, and copied credentials are useless elsewhere.
  17. PlannedAgent

    OS-native code signing

    Apple Developer ID signing and notarization, Windows Authenticode, and signed release tags. These are prerequisites for MDM package deployment and macOS privacy grants.

  18. PlannedAgentPlatform

    Enrollment v1

    One generic signed package per OS (.pkg, .msi, .deb/.rpm) plus single-use enrollment tokens, deployable by MDM, GPO or any package tool.

  19. PlannedPlatform

    Desired-state compliance

    Compliance checks and policies unified into one definition of "compliant" that is both measured and, later, enforced, with per-device evidence and a starter baseline.

    Starter baseline
    • Disk encryption, firewall, screen lock, OS updates within N days, EDR running, agent current
    • USB mass storage disabled or encrypted-only; no personal cloud-sync clients
    • Removable-drive encryption; MDM enrolled (macOS)
  20. PlannedAgent

    Reliability & project health

    Rollback when an update installs but won't start, coverage to 90%, OpenSSF Best Practices silver, SPDX headers.

  21. Next · Q1–Q2 2027
  22. PlannedAgentPlatform

    Enforcement with drift correction

    On Windows and Linux: if a setting drifts, it is detected and put back within one cycle, rolled out in rings, with a kill switch per check.

  23. PlannedPlatform

    Native macOS MDM pilot

    Apple's built-in MDM, served by our own backend: FileVault with key escrow, passcode and firewall profiles, silent privacy grants for the agent, and agent deployment through MDM.

    Why MDM, and why macOS first
    • Since macOS 11, configuration profiles can't be installed silently except through MDM
    • On Windows and Linux the agent already enforces settings directly
    • Built in our existing TypeScript backend, with a permanent device-facing hostname
  24. Later · H2 2027
  25. PlannedPlatform

    MDM at scale

    Zero-touch enrollment through Apple Business Manager, Declarative Device Management for OS updates, hardware-attested device identities, and lock/erase behind just-in-time approval.

  26. PlannedAgent

    DLP visibility

    USB and removable-media events, print auditing, and an opt-in data-at-rest scan that reports types and counts, never content.

  27. Planned

    Independent security review

    An external review of the agent, updater, release pipeline and device-facing backend, plus a second maintainer for the agent repository.

  28. Parked, on purpose
  29. Parked

    Content-aware DLP, kernel drivers, browser extension

    High cost and fleet-wide crash risk, and they would break the single static binary. Read why ↓

Decisions, in the open

What we build next, and what we won't.

A roadmap is only as good as the things it says no to. Both of these decisions are written up in full on the project wiki.

● Building next

Native MDM for macOS

On Windows and Linux, an agent running as SYSTEM/root can enforce policy itself. macOS routes that authority through Apple's MDM protocol, so we're adding it to the backend we already run.

Platform
Agent can enforce
Needs MDM
Windows
Mostly yes
Optional
Linux
Yes
No standard
macOS
Mostly no
Yes

Read the MDM assessment →

● Deferred

Content-aware DLP

Most of DLP's real-world value comes cheap: USB control, encryption and posture, which are already on the roadmap. Real-time content blocking needs kernel drivers on Windows and macOS. That cost is high and the return is low.

0 · Posture
DOING
1 · Visibility
LATER
2 · OS controls
DOING
3 · Blocking
PARKED

Bar length = relative engineering effort.

Read the DLP assessment →

Get started

Verify, install, done.

Every device installs from a device bundle downloaded from your console. It holds the agent, a config.json with that device's credentials, the installers and SHA256SUMS. The installer puts the agent in place, installs the config and starts the service.

  1. Download the bundle

    In the console, open the agent download, pick the OS and CPU architecture, and unzip the bundle on the device.

  2. Verify it

    Check every file against SHA256SUMS. The installers also verify the agent binary themselves, and refuse to run without the checksum file.

  3. Run the installer

    Use the terminal steps below, or double-click install.command (macOS), install.bat (Windows) or sentinelgo-install.desktop (Linux).

# In the unzipped device bundle
$ sha256sum -c SHA256SUMS --ignore-missing

# Installs to /opt/sentinelgo, adds config.json, starts systemd
$ chmod +x install.sh
$ sudo ./install.sh

# Confirm the service is running
$ sudo ./install.sh status
# In the unzipped device bundle
$ shasum -a 256 -c SHA256SUMS --ignore-missing

# Picks the right binary for this Mac, installs to
# /opt/sentinelgo, adds config.json, loads launchd
$ chmod +x install.sh
$ sudo ./install.sh

# Confirm the service is running
$ sudo ./install.sh status
# In the unzipped device bundle; compare each hash
# with its line in SHA256SUMS
PS> Get-FileHash .\install.bat -Algorithm SHA256
PS> Get-FileHash .\sentinelgo-windows-amd64.exe -Algorithm SHA256

# From an elevated prompt (or double-click: it asks
# for admin rights). Installs the agent and config.json
# and starts the Windows Service.
PS> .\install.bat

Updating, uninstalling, provenance checks and troubleshooting: INSTALLATION.md

See what it collects, first

Any release binary prints everything it would report, without a config, uploading nothing:

./sentinelgo-linux-amd64 -debug-dump

Build it yourself

Requires Go 1.26+. Builds are reproducible, so you can compare your binary with the published checksums.

git clone https://github.com/BrainStation-23/SentinelGo.git
cd SentinelGo
make build        # -> bin/sentinelgo
make test
make verify-cross # every OS/arch, cgo off
FAQ

Questions, answered straight.

Is SentinelGo an MDM?

Today it is agent-based endpoint management: inventory, posture, CVEs, compliance, policies, workflows, software deployment and remote actions on Windows, macOS and Linux. On Windows and Linux the agent can apply settings directly.

Native Apple MDM is on the roadmap as the next major step, because macOS only allows certain controls (configuration profiles, privacy grants, FileVault escrow) through MDM.

Does it install a kernel driver or system extension?

No. The agent is a user-space service written in pure Go with cgo disabled. That's a deliberate rule enforced in CI: it keeps the binary static, cross-compilable and free of the crash risk that comes with kernel code.

What data does the agent collect?

Hardware and software inventory, OS services, network configuration, security posture and OS audit logs. It does not collect credential material, and it redacts credential-like and PII patterns from task output before upload.

Run -debug-dump to print everything the agent would send, with nothing uploaded.

How are updates kept safe?

Releases are built reproducibly, signed with Ed25519, listed in SHA256SUMS, and shipped with provenance attestations and an SBOM. The agent verifies the signature and checksum before an atomic swap, and refuses unsigned binaries.

Which platforms are supported?

Windows (amd64), macOS (Apple silicon and Intel) and Linux (amd64 and arm64), running as a Windows Service, launchd daemon or systemd unit.

How do I report a security issue?

Please report it privately through the security policy, not as a public issue.

Help us build the endpoint agent you'd actually trust.

The agent is open source under Apache 2.0. Bug reports, feature requests, docs and code are all welcome.

Star on GitHub Contribute
Report a bug →

Found something broken? Use the bug report template.

Request a feature →

Tell us what would make SentinelGo more useful for your fleet.

Read the wiki →

Roadmap, design decisions, security process and how we work.