Air-gapped runtime.
Audit-grade evidence.

"Walk us through the code path. We need file:line."

SOC 2 CC6.1: "Logical access security software, infrastructure, and architectures over protected information assets have been implemented..." — your auditor's follow-up: *"show me the code path every authenticated user traverses to reach PII."*

PCI-DSS 6.5: "Address common coding vulnerabilities..." — auditor's follow-up: *"show me where you check for injection in every endpoint that touches cardholder data, and prove untrusted input can't reach a sink without sanitisation."*

HIPAA §164.312(a)(1): "Implement technical policies and procedures..." — auditor's follow-up: *"prove every PHI read goes through the audit-log function."*

These are reachability questions. Card 2 produces the answers as structural reachability proofs — graph paths from source to sink, with file:line citations every auditor accepts. Pairs with Vanta / Drata / SecureFrame (which track policies + checkboxes; ArgosBrain walks the call graph).

🔴 Nine perspectives. Compliance-mapped output.

/argos-security runs nine attacker perspectives + a Chain Composer that produces MITRE ATT&CK + Cyber Kill Chain + Unified Kill Chain phase mappings. Findings ranked by (impact × confidence) / cost-to-exploit. Each chain carries:

• Per-step file:line citations
• Pre/postcondition lattice (auditor-readable state transitions)
• MITRE ATT&CK technique IDs (CISO board-deck-ready)
• Historical-campaign analogies (CVE-2024-3094, log4shell, etc.)
• Severity bucket (critical_reachable / high / medium / informational_dead_code)

For SOC 2 evidence packs specifically: /argos-soc2-prep (full pack opt-in) produces a 13-bucket evidence pack mapped 1:1 to AICPA Trust Services Criteria 2017 (CC1-CC9 + A/C/PI/P). SOC 2 detail →

🛡 Stop the bug before it lands.

Audit-time evidence is downstream. Upstream fix: every developer commit goes through hooks + rules that enforce structural review before merge. Card 1 makes preflight + verify_no_fake_done mandatory pre/post every Edit, deterministic in Claude Code via hooks, ~90-95% compliant in Cursor / Aider / Cline via rules.

For regulated environments where every change must be traceable: hooks + rules log every preflight to ~/.argosbrain/hooks.log (rotatable, SIEM-shippable in the Custom Binary tier). Auditor can replay: *"on every Edit to a PII-handling function, was structural review performed?"* — yes, here's the log.

Local-first. Always. Egress promise.

Ingestion, storage, retrieval, hooks — all on your machine. Brain lives at ~/.argosbrain/brain.bin. The engine does not call out for retrieval. The dashboard runs on 127.0.0.1:3733. Skills compose tools via the in-process router.

The only non-local traffic is a periodic license check-in (~1 KB, daily), and only on installs with a license key. Free tier doesn't need one. The license check-in payload carries: license UUID, machine fingerprint hash, ArgosBrain version, OS family. It does NOT carry any code, symbol names, brain content, or telemetry on tool usage. That's the Egress Promise; it's enforceable.

For Custom Binary tier: fully offline-vendored — no license check-in, no network calls of any kind. We hand-deliver the binary + a custom license file with a 10-year validity. You can deploy to an air-gapped DMZ and never see another packet leave.

What's in the paid tier.

Solo + open-source is free, unlimited, forever (license-gated only for abuse prevention; no feature gating). Team / Custom Binary unlocks:

• SSO integration — SAML / OIDC for the management dashboard
• Per-team RBAC — fine-grained role policies on dashboard + brain access
• Audit-log SIEM export — JSON-lines stream of every Argos call, file:line citation, agent identity, timestamp. Wires to Datadog / Splunk / Sumo / Elastic / your-pipe-of-choice.
• On-prem deployment — fully air-gapped, no internet at all. Custom license file with 10-year validity. Bring-your-own SCIP/SCIPy/SCIP-go indexers if your registry is private.
• Custom skill catalogue — your team's internal "we always run this skill before merging X" workflows, bundled as part of argosbrain init.
• Priority support — direct Slack / email channel to engineering; response SLA in business hours.

What we don't promise.

ArgosBrain is structural reachability. We provide necessary-but-not-sufficient evidence: a graph path from source to sink. We do not prove tainted data actually flows (a sanitiser on the path may neutralise the risk; we report the path, not the dynamic taint).

Cloud IaC files are read as text. We do not parse Terraform / CloudFormation / Pulumi ASTs. Deep cloud audits need Prowler / Checkov / Pacu in addition. Same for dynamic analysis: pair with Semgrep diff-mode, CodeQL, or your runtime SAST.

The Chain Composer score is heuristic. Use it as triage signal, not as legal-grade exploitability proof.

These limits are documented inside every Card 2 report so your auditor sees them up front. The honest-about-limits posture is itself often a positive signal in regulated audits.

Talk to engineering, not sales.

[email protected] → · Compliance evidence patterns · SOC 2 evidence pack · Security disclosures + Egress Promise · The two services