Culvii Kit · The SDK for enterprise AI agents

Primitives

Five primitives. Infinite agents.

Culvii Kit is built around five composable primitives. Each one maps to a real engineering concern — not an academic abstraction.

A

Actor

The identity-bearing unit of execution. Wraps a model, tools, and capabilities into a single accountable entity.

actor.create()

W

Workflow

Composable execution graph. Sequential, parallel, and conditional paths with built-in checkpoints.

workflow.run()

T

Tool

Typed, capability-scoped function execution. Every call is logged and signed.

tool.define()

H

HITL

Human-in-the-loop checkpoint. Suspends execution, awaits approval, resumes.

hitl.await()

M

Model

Provider-agnostic LLM configuration. OpenAI, Anthropic, Azure, or local.

model.configure()

Agent Identity

You wouldn't deploy a microservice without authentication. Don't deploy agents without identity.

Every agent you deploy with Culvii Kit is assigned a DID backed by Ed25519 keypairs. Every action an agent takes is signed with its private key.

Your security team can verify any agent's identity across the organization — even across tenants — without a central authority. Compromise one agent; don't compromise them all.

Read the identity docs

identity.ts

copy

import { Actor, Model } from '@culvii/kit'

const agent = new Actor({
  name: 'lease-extractor',
  model: Model.openai('gpt-4o'),
  capabilities: ['read:leases'],
})

// DID issued · Ed25519 keypair generated
console.log(agent.did)
// → did:culvii:jll-prod:lease-extractor-a3f9b2

// Every action is signed
await agent.run({ prompt: 'Summarize Q2 leases' })
// ✓ signed · ✓ logged · ✓ hash-chained

Capability Governance

Your CISO's biggest fear isn't AI failing. It's AI succeeding at the wrong thing.

Culvii Kit's capability registry lets you define a precise trust surface for every agent. What data it reads. What systems it writes to. Which other agents it can delegate to.

Trust elevation requires explicit grants. Capabilities expire. Keys rotate without downtime. Revocation is instant. This is what enterprise governance looks like for AI agents.

Read the governance docs

capability registry · lease-extractor registry.acme-prod · v4

read:leases 90d · q2-portfolio

tool:extract_clauses rate · 100/h

delegate:valuation-writer hitl-required

write:valuations denied

read:client-pii denied

grants · 3 of 2,847 ambient authority · 0

Immutable Observability

Log files lie. Hash chains don't.

Every observation event emitted by your agents flows through a hash chain buffer into a Merkle Tree. The result is a cryptographically verifiable audit log — not a file that can be silently edited after the fact.

When regulators ask "what did your AI agent do on March 14th at 2:47 PM," you can prove it — with cryptographic certainty. Not a screenshot. Not a log file. A tamper-evident chain of evidence.

Read the observability docs

merkle tree · verified depth · 3 · 8 events

root · f4a2…8b91

h01 · c8d1…

h02 · 9e4a…

tool.call

model.inv

hitl.apv

tool.call

events · 142 · signed tampered · false

Capabilities

Everything you need. Nothing you don't.

Cryptographic Agent Identity

DID-based identity. Ed25519 keypairs. Verifiable, revocable, portable across tenants.

Capability Registry & Governance

Precise trust surface per agent. Grant, scope, rotate, revoke. No ambient authority.

Immutable Audit Trails

Hash-chain + Merkle Tree observation events. Cryptographically verifiable.

Human-in-the-Loop (HITL)

Approval workflows built into execution. Slack, email, or webhook channels.

Multi-Tenant Auth

Per-tenant key scoping, rotation without downtime, per-key rate limiting.

TypeScript + Python SDKs

Strongly typed, fully IntelliSense-supported, provider-agnostic for any LLM.

Why Culvii Kit

Not another agent framework.

Capability	Culvii Kit	LangChain	CrewAI	Build yourself
Cryptographic agent identity (DID)	✓	—	—	Custom build
Capability registry & governance	✓	—	Partial	Custom build
Immutable audit trails (Merkle Tree)	✓	—	—	Custom build
Human-in-the-loop checkpoints	✓	Plugin	Plugin	Custom build
Multi-tenant key management	✓	—	—	Custom build
TypeScript + Python first-class SDKs	✓	✓	✓	—

The SDK for enterprise AI agents your security team will approve.