Shiva | Autonomous AI Governance Protocol

Your AI agents are only as safe
as the system governing them.

Shiva is a Zero-Knowledge, Stateless Agent Guardrail — it sits between your agents and the world. Every action is evaluated before it runs, unsafe ones are blocked, and every decision is sealed into a tamper-proof cryptographic audit chain.

"Your AI agents are only as safe as the system governing them."

"If your agent can break Shiva, it better be good at breaking Bitcoin."

Live challenge · runs in your browser

Set a rule for your agent. Then try to break it.

You play the rogue agent — slip a violation past Shiva if you can.

1 · The rule Shiva enforces

2 · What your agent is about to say or do

0 sent · 0 blocked · 0 allowed

{ "verdict": "BLOCK", "reason": "Governance rule triggered: High Risk Activity", "evaluation_id": "e3b0c442-98fc-1c14-9afb-4c8996fb9242", "chain_position": 1042, "detection_method": "keyword", "latency_ms": 3, "cached": false }

1,000 evals each. Resets every 12 hours.

While we launch, every agent gets 1,000 evaluations free. As an agent runs low you'll see it in the dashboard. When it's used up, that agent's governance pauses until its 12-hour window resets — then it's free again. No card, no charge — fail-safe by design.

✓

Running. Up to 1,000 evals per agent, across 3 agents — fully free.

Low. At 20% remaining the dashboard flags the agent.

■

Used up. That agent pauses (HTTP 429) and auto-resumes in 12 hours. Metered plans take effect August 2026.

I can't really code. I built this anyway.

I'm not an engineer. I built Shiva by sitting with AI — talking the idea through across different AI platforms, and finally bringing it together with Claude.

The plan was to build something small. One of those AIs told me I'd set out to build a bicycle to ride to the store for milk — and somehow ended up building a spaceship. That's honestly how it felt.

Before this I was a chef, happily running a little café in Rishikesh, India. Today I'm up in the hills, building software to solve real problems — and Shiva is the first one I couldn't stop building.

The cryptographic anchor at the core of every sealed verdict is signed DKBK — my initials. It's literally my name holding the chain together. If you're going to trust Shiva with your agents, I think you deserve to know the person behind it.

🤖 Your agent
takes an action ›

⚡ Shiva checks
if it's safe ›

✅ You get a
signed verdict

Free during launch — up to 3 agents · 1,000 evals each · no credit card

Keep your AI agents safe.

Add tamper-proof governance to any AI agent in minutes.

Initialize an isolated governance partition in the Shiva kernel — SHA-256 audit chain, HMAC-sealed verdicts, per-tenant advisory locks.

Company or Project Name

Work Email

Password

Paid plans — coming soon (preview only)

Pay-As-You-Go · $5 once · 25k evals Developer · $19/mo · 15k evals/day Scale · $59/mo · 50k evals/day Production · $149/mo · 150k evals/day Enterprise · $500/mo · 750k evals/day

Already have an account? Sign In

← Back to homepage

Step 1 of 2 — Your API Key

🔑 Save your API key.

Your agents use this as the X-API-KEY header on every call to Shiva. We only show it once — copy it now. If you lose it, rotate it from Settings.

Master authorization token — injected as X-API-KEY on all agent evaluation requests. Hashed at rest (bcrypt). Issued once; rotate via POST /auth/rotate-key.

X-API-KEY

sh_live_loading...

Step 2 of 2 — Audit Delivery

📬 Where should we send your audit logs?

Shiva does not store your audit data. Every sealed verdict is streamed straight to you — you own it. Without a webhook, those records are gone. Add one below so nothing gets lost.

Configure redundant HTTPS endpoints to receive HMAC-signed cryptographic audit events. The governance kernel streams sealed chain blocks to your infrastructure via the webhook worker in real time — zero retention on our side.

🤖 Your Agent calls /evaluate

→

⚡ Shiva seals & signs

→

🏦 Your Webhook = your storage

Primary Endpoint Required

Backup Endpoint (optional — failover)

⚡ Need a webhook endpoint fast? Try one of these:

webhook.site — instant, free Pipedream — free + searchable logs RequestBin — inspect each payload

👆 Go to one of these, grab your unique URL, paste it above. Takes 30 seconds.

Total Evaluations

All time

Monitored Agents

Registered and active

Active Rules

Governance policies running

Audit Chain

Secure

100% Verified

Recent Governance Decisions

Time	Agent	Verdict	Reason	Block #
🤖 No evaluations yet Register an agent, then call `POST /evaluate`. Verdicts will appear here in real time.

🔒 Shiva Trust Seal

Every verdict is SHA-256 linked to the previous block. Tamper any record — the chain breaks instantly.

DKBK cryptographic anchor — SHA-256 hash chain + HMAC, key_id versioned, per-tenant pg_advisory_xact_lock. Hash verified against stored canonical payload_text.

Chain Status

✅ SECURE

All blocks sealed correctly

Agent ID	Name	Type	Status	Registered
🤖 No agents yet Register your first agent below to start governing its actions.

Rule	Category	Severity	Status
📋 No rules yet Create a rule below. Example: "must not give medical advice" or "never share user passwords".

Integration

Your API Key

••••••••••••••••••••••••••••••••

Use this as the X-API-KEY header in every agent call to Shiva. Lost it? Rotate to issue a new one — the old key stops working immediately.

🧪 Test an Evaluation

Simulate an agent action and see what Shiva decides — great for testing your rules.

Agent Name

What the user sent to the agent

What the agent was about to reply

🤖 Governance AI Engine

Choose which AI model judges your agent's actions. Bring your own key — you pay your provider directly, not us.

BYOM (Bring Your Own Model) — cascade: keyword → gemma2 (Ollama) → Haiku, escalating only when unsure. BYOM keys Fernet-encrypted via BYOM_ENC_KEY (MultiFernet, rotatable). Circuit breaker on all LLM providers.

Active Engine Shiva Default (gemma2:2b)

AI Provider

Model Name

API Key 🔒 Stored encrypted. Leave blank to keep the existing key.

API Base URL

📬 Audit Delivery Endpoints

Shiva streams every sealed verdict to your webhook in real time. Update the URLs here if they change.

Dual redundant HTTPS webhook targets. The webhook worker fans out HMAC-signed payloads to primary + failover endpoints. SSRF guard active — private/internal IPs blocked, redirects disabled.

Primary Webhook

Backup Webhook (optional — failover)

Your AI agents are only as safe
as the system governing them.

Set a rule for your agent. Then try to break it.

What happens at 3 a.m.
when your agent goes rogue?

Your agent asks. Shiva decides.
In under 200ms.

One call before each action.

Your data stays yours.
Your governance can't be touched.

Don't trust us. Recompute the chain.

One core. Everything around it.

Free while we launch.

Free Trial

Embed the kernel. Keep 70%.

1,000 evals each. Resets every 12 hours.

Questions, answered

I can't really code. I built this anyway.

Ship agents you can actually trust — and prove it.

Hit a snag, or have feedback?

🔑 Save your API key.

📬 Where should we send your audit logs?

👋 You're in! Here's what to do first.

Recent Governance Decisions

🔒 Shiva Trust Seal

Your AI Agents

Register an Agent

Governance Rules

Create a Rule

🔗 Sealed Audit Chain

Chain Verification

Integration

🧪 Test an Evaluation

🤖 Governance AI Engine

📬 Audit Delivery Endpoints

Your AI agents are only as safe as the system governing them.

Set a rule for your agent. Then try to break it.

The problem: what happens when an AI agent acts without authorization at 3am

What happens at 3 a.m.when your agent goes rogue?

The flow: how Shiva's governance kernel intercepts agent actions through four steps and maps each problem to a solution

Your agent asks. Shiva decides.In under 200ms.

One call before each action.

Why Shiva is different: your data stays in your storage, and every decision is hash-chained and HMAC-signed so a compromised agent can't forge or rewrite it

Your data stays yours.Your governance can't be touched.

Don't trust us. Recompute the chain.

One core. Everything around it.

Free while we launch.

Free Trial

Embed the kernel. Keep 70%.

1,000 evals each. Resets every 12 hours.

Questions, answered

I can't really code. I built this anyway.

Ship agents you can actually trust — and prove it.

Hit a snag, or have feedback?

Keep your AI agents safe.

Welcome back.

🔑 Save your API key.

📬 Where should we send your audit logs?

👋 You're in! Here's what to do first.

Recent Governance Decisions

🔒 Shiva Trust Seal

Your AI Agents

Register an Agent

Governance Rules

Create a Rule

🔗 Sealed Audit Chain

Chain Verification

Integration

🧪 Test an Evaluation

🤖 Governance AI Engine

📬 Audit Delivery Endpoints

Your AI agents are only as safe
as the system governing them.

What happens at 3 a.m.
when your agent goes rogue?

Your agent asks. Shiva decides.
In under 200ms.

Your data stays yours.
Your governance can't be touched.