Caisson vs Sentrik
Sentrik gates what gets written and governs the agent writing it. Caisson is the control that runs in production once it is written. Here is the honest line between a scanner and a substrate.
Which should you use?
Pick Sentrik to govern how code gets written: a language-agnostic CLI and CI gate that scans commits against 632 rules in 26 framework packs, binds an AI agent to a declared file scope and blocks it when it writes outside, and emits HMAC-signed per-commit attestations. Pick Caisson for what the code has to do at runtime: fail-closed Postgres RLS with isolation tests, a WORM + hash-chained audit trail, per-tenant field encryption, and evidence packs generated from the live system. Sentrik inspects the diff; Caisson is the control that holds in production. They stack rather than compete.
What Sentrik is
AI-code compliance scanner + agent-governance layer. Facts below were read from sentrik.dev on 2026-08-09.
- A CLI and CI gate (sentrik scan, sentrik gate, sentrik attest) distributed via PyPI, pip, and Docker, with a VS Code extension, a GitHub Action, and an MCP server for Claude Code, Cursor, and Cline (sentrik.dev, read 2026-08-09).
- 632 rules across 26 standards packs on the top tier — SOC 2, HIPAA, PCI DSS, ISO 27001, GDPR, EU AI Act, NIST 800-53, NIST AI RMF, CMMC 2.0, IEC 62304, DO-178C, ISO 26262, MISRA C/C++, 21 CFR Part 11 and more; the free tier is 6 packs / 193 rules.
- AI-agent governance: sentrik task-bind binds an agent's declared intent to a file glob, then flags and blocks writes outside that scope, with a signed incident record and one-click rollback.
- HMAC-SHA256-signed per-commit attestations, a token-gated read-only auditor portal, SBOM generation and CVE scanning against OSV.dev, SARIF/JUnit/CycloneDX output, and air-gapped offline license validation.
- Labelled 'Early Access'; the paid Team, Organization, and Enterprise tiers are contact-gated with no published numbers (checked 2026-08-09).
An honest comparison
Where Sentrik has a capability, it is marked. Caisson is the compliance and tenant-isolation substrate; Sentrik wins the rows it wins.
| Detail | Caisson | Sentrik |
|---|---|---|
| Form factor | Library composed into your app | CLI + CI scanner over your repository |
| Scans commits against framework rule packs | — | |
| Blocks non-compliant code in CI | — | |
| AI-agent task-scope binding + rollback | — | |
| Languages other than TypeScript | — | |
| Evidence signing | Ed25519 + RFC-3161 (public verification) | HMAC-SHA256 (shared secret) |
| Fail-closed Postgres RLS + automated cross-tenant isolation tests | — | |
| WORM evidence store + append-only hash-chained audit trail | — | |
| SOC 2 / HIPAA / EU AI Act evidence packs + OSCAL export | — | |
| Per-tenant field encryption (envelope, per-tenant key) | — | |
| Published price · one-time perpetual, own the source | Free tier; paid tiers contact-gated |
What Sentrik is genuinely better at
A comparison that only flatters one side isn't worth reading. Here is what this kit does well.
26 packs reaching into safety-critical regimes — IEC 62304, DO-178C, ISO 26262, MISRA C/C++, 21 CFR Part 11 — across multiple languages. Caisson covers a deliberately narrow set of regimes for TypeScript on Postgres; if you build medical-device, avionics, or automotive software, Sentrik is addressing a surface Caisson does not touch.
Task-scope binding with a signed incident record when an agent writes outside its declared globs is real AI-governance machinery, and it drops into an existing pipeline with no application changes — pip or Docker, a GitHub Action, and an MCP server the coding agent itself calls. That acquisition path is genuinely lighter than adopting a substrate.
Where Caisson draws the line
The compliance and tenant-isolation substrate a launch kit leaves to you.
Static analysis reports on code; it cannot make a query fail closed. Caisson's isolation is a Postgres FORCE RLS policy with a test that attempts a cross-tenant read and asserts it returns nothing, on every run; its audit trail is hash-chained rows anchored in S3 Object-Lock WORM. Those hold at 3am against a bug a scanner had no rule for.
Sentrik signs attestations with HMAC-SHA256, a shared-secret MAC: verification needs the same key that signed it. Caisson's evidence packs carry detached Ed25519 signatures plus RFC-3161 timestamps over WORM-retained objects, so a third party — an auditor, a customer, you after the vendor is gone — verifies them with a public key and no cooperation from us. Both are real signing; only one is externally verifiable.
Sentrik's paid tiers are contact-gated with no published number, and its free tier is a scanner subscription, not source you own. Caisson publishes a committed one-time price, ships an Apache-2.0 base, and hands you the code: nothing to renew for the controls to keep working.
Which should you pick?
You want language-agnostic rule coverage over AI-written code in CI — especially safety-critical regimes like IEC 62304, DO-178C, or MISRA — and a gate that blocks merges without changing your application.
You need the controls themselves to exist at runtime — fail-closed tenant isolation the database enforces, a WORM-anchored audit trail, and evidence generated from the live system — in TypeScript on Postgres, owned one-time.
They stack in the obvious order: Sentrik gates what gets written, Caisson is what runs. A scanner has less to find in a codebase whose controls are already implemented and tested, and the runtime evidence is the part an assessor actually tests.
One-time, own the source.
Caisson is a one-time perpetual license — the price never recurs, and it includes 12 months of updates from your purchase date, renewable per entitlement afterward at 40% of list per year. The Base substrate is Apache-2.0; the compliance modules are commercial.
Compliance bundle
$1,649, one-time. Fail-closed RLS, WORM, the audit chain, evidence packs, and the framework and signing carves — the whole substrate this comparison is about.
À la carte
Take a single module from $49 — audit-worm, field-crypto, or compliance-core on their own, onto your existing Postgres app.
Everything bundle
$2,259 covers every bundle and every à-la-carte module, plus the open base, in one purchase.
Common questions.
Is Caisson a Sentrik alternative?
Both produce signed attestations. What's the difference?
What does Sentrik cost?
Does Caisson govern AI agents the way Sentrik does?
Ship the compliant backend.
Explore the Compliance bundle, browse every module in the marketplace, or read another comparison.