TYPESAFE AI JEV · LIVE VIA VERCEL AI GATEWAY · v0.1.0

Verifiable evidence for Jev decisions.

Jev returns typed probabilistic decisions. Your application applies the policy. LoopGrid preserves what was decided, why it was allowed or escalated, what happened afterward, and the cryptographic evidence needed to verify the record.

View GitHub integration ↗View v0.1.0 release →

loopgrid-jev v0.1.0 · Live validated with typesafe-ai/jev through Vercel AI Gateway · LoopGrid Core 0.8.1-design-partner.

LIVE DECISION EVIDENCE FLOW

01State + typed Jev questions

02Live Jev answers + probabilities

03Application threshold / policy

04Signed decision + delegated authority

05Sandbox action + observed outcome

06Evidence complete + verify

Jev decides. LoopGrid preserves the evidence. The included refund action is sandbox-only; LoopGrid does not execute real refunds or payments.

01 · INSTALL

Clone the released integration and install the Python adapter.

git clone https://github.com/loopgridio/loopgrid-jev.git
cd loopgrid-jev
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -e .

The repository also includes a matching Node example. v0.1.0 keeps the Jev decision step separate from LoopGrid's evidence role.

02 · LIVE JEV ACCESS

Use the validated Vercel AI Gateway path.

ProviderJEV_PROVIDER=vercel

Jev modeltypesafe-ai/jev

Gateway credentialAI_GATEWAY_API_KEY from your Vercel AI Gateway account

LoopGrid credentialWorkspace service key with the minimum scopes needed by your flow

The release validation used the TypeSafe-compatible Jev path exposed through Vercel AI Gateway and received live Noul/Choice results from typesafe-ai/jev.

03 · LOOPGRID CONTRACT

Carry the live decision all the way to observed outcome evidence.

live Jev inference
  → typed answers + probabilities
  → deterministic application threshold
  → LoopGrid decision_created
  → LoopGrid model_completed
  → LoopGrid policy_evaluated
       auto_allowed | human_approval_required | blocked
  → delegated authority evidence
  → external application action
  → LoopGrid tool_executed
  → LoopGrid outcome_observed
  → evidence_complete
  → verify

The integration uses the same X-LoopGrid-Key authentication and native decision/event/review/verification contract as LoopGrid's other adapters. It does not introduce a Jev-specific ledger or execution engine.

04 · SAMPLE POLICY

Jev supplies uncertainty. Application code owns the boundary.

The refund example asks a Noul question for whether a refund is requested and a Choice question for operational risk. The sample routes:

01Auto candidate

refund ≥ 0.90, risk is low and Choice confidence is at least 0.70 → LoopGrid auto_allowed.

02Human review

Ambiguous cases → LoopGrid human_approval_required and the native review lifecycle.

03Blocked

refund < 0.50 → LoopGrid blocked.

These thresholds are example application policy, not universal recommendations. Calibrate them for your own labelled data, authority limits and risk model.

05 · RUN LIVE

Run the same live path used for the v0.1.0 release gate.

$env:JEV_PROVIDER="vercel"
$env:AI_GATEWAY_API_KEY="<your Vercel AI Gateway key>"
$env:LOOPGRID_BASE_URL="http://127.0.0.1:8000"
$env:LOOPGRID_WORKSPACE_ID="default"
$env:LOOPGRID_API_KEY="<your LoopGrid service key>"
python examples\refund_gate.py

The included external refund service is deliberately a sandbox demo. Its recorded action/outcome includes sandbox: true and real_money_moved: false. Your production application remains responsible for any real business action.

06 · RELEASE VALIDATION

The released path reached complete evidence and valid verification.

Live modeltypesafe-ai/jev through Vercel AI Gateway

Lifecycleevidence_complete

Evidence coverage100% in the validated sandbox action/outcome path

Verificationvalid: true · failures: []

Automated tests21 tests in the v0.1.0 release repository

LoopGrid verification establishes integrity and provenance of what was captured. It does not prove that Jev's judgment was correct, that every upstream statement was truthful, or that a real external refund occurred.

The public example intentionally records a sandbox action and observed outcome so the evidence lifecycle can be demonstrated without moving real money.

Independent community integration; LoopGrid is not affiliated with or endorsed by TypeSafe AI or Vercel.

JEV + LOOPGRID

From live judgment to portable proof.

Use Jev for the typed decision, keep policy and execution in your application, and preserve the resulting decision-to-outcome evidence in LoopGrid.

Open integration →All integrations