One platform. Eight surfaces. The full Safeguard ecosystem.
The spine is the same everywhere: three model families, eleven scanners, one policy engine, one audit log. The surface area is how engineers actually touch it — models served as an API, a desktop app, a local coding agent, IDE extensions, a CLI, an MCP server, the portal, and a marketplace. Eight ways in, one source of truth.
Every way you touch the platform.
Each surface is its own product, but they all read the same policy, write the same audit log, and share the same SBOMs. Eight entry points, one substrate.
Models-as-a-Service
Griffin · Eagle · Lion served behind a stable REST API.
The three model families served behind a stable, versioned REST API. Pay per inference, or bundle into a deployment tier. The auto-router decides which Griffin variant earns the workload — Lite, S, M, L, or Zero — based on a complexity score Eagle assigns up front.
- Single endpoint, automatic variant selection
- Per-tenant inference isolation; no cross-tenant KV cache reuse
- Streaming, batched, and async inference modes
Desktop Application
Native app for Mac, Windows, Linux. The full local control plane.
Native desktop app for Mac, Windows, and Linux. Takes full control of the local system — scans repos, runs CI-equivalent checks, executes auto-fix plans, and watches in the background for drift. Built for the engineer who wants the platform on their own machine, not behind a tab.
- Background drift watcher with notifications
- Local CI-equivalent gate before push
- Available via safeguard-sh.itch.io/safeguard
Safeguard Code
A local AI coding agent that already knows your supply chain.
Local AI coding agent that runs on your system, similar in spirit to Claude Code. Drives the editor, the file system, the build, and the test runner. Lion handles the hot path on-device with sub-100 ms latency; Griffin cloud-bursts for the genuinely hard reasoning that needs a real budget.
- Terminal-first agent, drives editor and shell directly
- Supply-chain-aware: reads policy, SBOM, reachability
- Auditable per-session trace, exportable for review
IDE Extensions
VS Code, JetBrains, Cursor. Lion at sub-100 ms.
First-party extensions for VS Code, the JetBrains family (IntelliJ, PyCharm, GoLand, WebStorm) and Cursor. Lion runs on-device for inline findings with sub-100 ms p95 latency; Griffin is available on demand when the inline answer needs deeper reasoning.
- Inline sink and sanitiser checks while you type
- Hover-tooltip enrichment: CVE, EPSS, KEV
- Quick-fix suggestions with PR-comment prefetch
CLI Tool
scan · sbom · fix · policy. Scriptable, air-gappable.
Eleven scanners and seven enrichment feeds behind four commands: safeguard scan, sbom, fix, policy. Output as JSON, SARIF, or CycloneDX. Pipes cleanly into any CI runner and ships as a static binary for environments that don't tolerate runtimes.
- @safeguard-sh/cli on npm, or a static binary
- Air-gapped mode with offline vuln DB bundle
- Same engine as the cloud platform, deterministic verdicts
MCP Server
Model Context Protocol with capability scoping and egress guards.
Model Context Protocol server with per-tool capability scoping, sensitive-data egress guardrails on Lion's egress path, and a prompt audit log with cryptographic chain-of-custody. Drop-in for Claude Code, Cursor, Cline, and any custom agent that speaks MCP.
- Tool allowlist plus per-agent capability bounds
- Lion-screened egress for PII, secrets, proprietary code
- Replayable session audit for incident response
Portal & Dashboard
Web console for security and engineering leadership.
The web console where the org actually reads its supply-chain posture. Trend lines, SLA tracking, regulator export, evidence store. Every finding from every other surface lands here with its provenance intact — one source of truth, not one tab per tool.
- Trend, SLA, and exception tracking by team
- One-click regulator export with chain-of-custody
- Evidence store: SBOMs, attestations, scan history
Marketplace
Curated integrations, SBOM bundles, compliance packs.
Curated integrations for Jira, ServiceNow, Slack, Splunk, and Datadog. Verified SBOM bundles, compliance pack downloads, and customer-built workflows that drop into the platform without a services engagement.
- Ticketing, alerting, and SIEM connectors
- Verified third-party SBOM bundles
- Customer-built workflow library
Different surfaces. One spine.
Every surface reads the same policy, writes to the same audit log, and shares the same SBOMs. Findings move sideways without translation — the editor that surfaced a sink and the dashboard that reports the SLA are looking at the same row.
Inline findings, inline fixes, inline policy. No round trip to the cloud for the hot path.
Repo-wide sweeps, reachability ranking, deep reasoning on the survivors. The same engine, scriptable.
Findings flow through the MCP and integration mesh into the central console. Leadership reads one view.
Pick the surface that matches the workflow.
Every surface, at every isolation level.
Pick the shape that matches your trust boundary; the surfaces stay consistent across them.
Shared cloud
Multi-tenant inference plane, fastest onboarding. Per-tenant isolation at prompt and KV cache layer.
Dedicated cluster
Single-tenant inference on isolated hardware. Deterministic latency, SHA-pinned weight attestation.
VPC-isolated
Customer-controlled VPC, BYO-KMS. The inference plane sits inside your network boundary.
Air-gapped / sovereign
No internet egress, on-prem GPU, full audit-log export. Suits regulated and classified deployments.
Single-purpose, on purpose.
Software supply chain security is the whole product. The corollary is what we don't ship.
Out of scope, by design
A general-purpose code assistant. Safeguard Code reasons about supply chain — not poems, marketing copy, or Pokédex queries.
An image generator, a chat companion, or a voice agent. The product is single-purpose by design.
A model trained on your code. Customer source code does not enter training under any tier, contractually and by audit.
A 'security copilot' draped over a generic LLM. The Griffin family is weighted for cybersecurity end to end, not prompt-engineered.
Pick a surface. Start anywhere.
The eight surfaces converge on the same audit log. Begin in the editor, the terminal, or the dashboard — your team will meet you at the others.
Self-healing security runs on Safeguard.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.