Mindburn Labs
DEMO
APACHE 2.0

The open-source safety engine for AI

HELM OSS checks every action your AI wants to take, applies your rules, blocks anything unsafe, and saves a proof record. It works with any AI framework. It's free and open-source.

View on GitHub Read the Docs
ARCHITECTURE

How the engine works

HELM OSS is the safety engine inside the HELM stack. It checks every action your AI wants to take, applies your rules, and creates a proof record. Click any stage to see what it does.

External BoundaryAgent sends structured proposals — never raw execution
Kernel — Fail-ClosedEvery proposal is validated, policy-checked, then executed under governance
Proof SurfaceReceipts chain into an offline-verifiable evidence bundle
Hover or click any stage to see its invariant and proof output
ExternalKernel boundaryProof surface
Full architecture docs
INPUT → GOVERNANCE → PROOF

From proposed action to verifiable proof

HELM turns policy inputs into deterministic execution records and offline-verifiable evidence without letting raw tool calls slip around the boundary.

Fail-closed runtimeOffline verificationMCP-native
HELM OSS execution latticeTyped policies, regulations, and permissions flow into the HELM core engine, which executes inside a fail-closed sandbox and emits audit trails, cryptographic proofs, and deterministic action records.POLICY SURFACESPROOF SURFACESTYPED GUARDRAILSPOL-01PoliciesJURISDICTION RULESREG-02RegulationsHUMAN APPROVALSPERM-03PermissionsFAIL-CLOSED KERNELHELMCore execution authorityCPI • PEP • WASI SANDBOX • RECEIPTSAudit TrailsCrypto ProofsCompleted Actions
Inputs arrive as typed constraints, not raw execution.HELM enforces policy before any side-effect is allowed.Outputs exit as receipts and proofs that can be checked offline.
Input Surface

Policies, rules, approvals

Governance starts with typed constraints from policy files, regulatory rules, and human permission checkpoints.

Kernel Boundary

Deterministic execution

The core validates, policy-checks, executes in a sandbox, and fails closed whenever the boundary cannot prove it is safe.

Proof Surface

Artifacts anyone can inspect

Every action leaves an auditable trail: receipts, proof structures, and completed action records that survive outside the runtime.

QUICKSTART

Set up in 5 minutes

HELM works with any AI framework. Here are the exact code changes for popular tools.

1
Clone the repo
$ git clone https://github.com/Mindburn-Labs/helm-oss.git
2
Build the engine
$ cd helm-oss && go build ./cmd/helm-proxy
3
Start the proxy
$ ./helm-proxy --config config.yaml
4
Point your agent
$ export OPENAI_BASE_URL=http://localhost:8420/v1
Full quickstart guide
MCP ECOSYSTEM

Works with every MCP client

HELM is MCP-native. Any client that speaks the Model Context Protocol can use HELM as its execution authority — no adapters, no shims.

FRAMEWORK COMPATIBILITY

Framework-agnostic by design

Change one URL — every framework gets receipts, policy enforcement, and offline verification.

Full compatibility matrix
OSS vs PLATFORM

What the engine includes. What the platform adds.

HELM OSS includes the full safety engine. HELM Platform adds team dashboards, approvals, compliance workflows, and multi-org controls on top of the same engine.

OPEN SOURCE

HELM OSS Kernel

$0
  • Checks every AI action against your rules. Blocks unsafe actions automatically.
  • Each decision is linked to the previous one, creating a tamper-proof chain.
  • Set cost and time limits for each action to prevent runaway spending.
  • Require a human to approve risky actions before they happen.
  • Built-in test suite to verify your setup works correctly.
HELM PLATFORM

Commercial control plane

Custom
  • Share safety rules across organizations with revokable access.
  • Compare AI model performance and costs across providers.
  • Plug into Salesforce, SAP, ServiceNow, and more.
  • Manage AI agents across multiple teams from one dashboard.
  • Automate SOC 2, HIPAA, and ISO 27001 compliance.
  • Run safety checks inside tamper-proof hardware for maximum security.
FAQ

Common questions

Add AI safety today

Start with the free engine. Add team features only when you need them.

View on GitHub Full Documentation