Dragons

Documentation

Build with
the trust layer.

Everything you need to govern your autonomous agents: identity, authorization, evidence, and liveness. From first dock to production fleet.

01

Getting started

Install the SDK, configure your first tenant, dock an agent, and observe a signed receipt in the WorkLedger. Five minutes from zero to proof.

Install

Add the Dragons SDK to your project and configure credentials.

Quick start

Dock your first agent, issue a lease, and produce a receipt.

Side-by-side mode

Run Dragons alongside your existing stack without changing agent execution.

02

Concepts

The four mechanisms that close the trust loop: cryptographic identity, signed authorization, hash-chained evidence, and organism-state liveness.

Identity & manifest hash

SHA-256 fingerprint of agent code, config, and capabilities. The cryptographic anchor.

Leases & authorization

Time-bounded, scope-restricted grants. No lease, no execution.

WorkLedger & evidence chain

Signed receipts, hash-chained, tamper-evident. Third-party replayable.

Organism state & liveness

Running, degraded, autohealing, quarantined, stopped. Compulsory heartbeat.

Tenants & docking

Multi-tenant isolation. Agents dock into a tenant with an authorized scope.

03

API reference

Complete reference for the Dragons HTTP API and SDK. Types, endpoints, error codes, and examples.

Authentication

API key management, tenant scoping, and key rotation.

Agents

Register, dock, query, and revoke agent identities.

Leases

Issue, renew, revoke, and inspect authorization leases.

WorkLedger

Append receipts, query chains, export evidence packs.

Heartbeat

Emit pulses, query organism state, configure thresholds.

04

CLI

The Dragons CLI for local development, CI pipelines, and fleet operations. Inspect, replay, and audit from the terminal.

dragons dock

Register and dock an agent from the command line.

dragons lease

Issue, inspect, and revoke leases.

dragons ledger

Query the WorkLedger, replay chains, export packs.

dragons status

Inspect organism state and heartbeat history.

Concepts

The four mechanisms.

The trust loop closes around four primitives. Each one has a dedicated concept page; the product page shows the same mechanisms as shipped surface.

FAQ

Questions buyers ask.

The objections that come up in every evaluation, answered against the actual mechanism. If a question is missing, email hello@dragons.dev.

We already use Temporal — why do we need Dragons?

Temporal tracks workflow state — whether the function ran and what it returned. Dragons tracks who the agent is (manifest_hash), what it was authorized to do (lease), what it proved it did (signed receipt), and whether it is still alive (organism state). The concerns are orthogonal; run both.

Couldn’t we build this ourselves with Postgres and a cron?

You could. The hard parts are lease signing under a key-rotation policy, hash-chained receipts that survive a third-party replay, and tenant isolation that holds up under audit. Those are the parts that quietly break six months in. Bring your watchdog code and we will show you what it misses.

Our stack is Elixir / Python / Go / TypeScript — does this support us?

You operate Dragons, you do not write it. The operator surface is a Go CLI and an HTTP API. Agents enroll via a thin SDK call in your language. The control plane itself is a black box you run; the language inside it is not your problem.

What is the license? Our legal team will ask.

We are finalizing the canonical license statement before launch. The current claim on the pricing page is a placeholder; the authoritative statement will live here once it is settled. If your procurement needs an answer before then, email legal@dragons.dev and we will respond directly.

How is this different from LangSmith or Datadog?

LangSmith traces what a model emitted. Datadog tracks CPU, memory, and request latency. Dragons tracks the four trust mechanisms above the model and below the infra: identity, authorization, signed evidence, and organism-state liveness. The receipts are designed to be replayed by a third party without trusting us or your tenant.

What is the smallest credible adoption — can we trial with one agent?

Yes. Side-by-side mode docks a single agent, issues one lease, and writes receipts to a tenant-local ledger without changing how the agent executes. Run it on one workload for thirty days, export the evidence pack, and read what it caught. The quickstart below is the literal path.

How do we know it scales to our fleet size?

Tell us your agent count, action rate, and receipt-retention window and we will give you a direct answer instead of a benchmark slide. Heartbeat cadence, lease issuance throughput, and ledger append rate are the three numbers that matter — ask with your workload and we answer with those.

Need help?

Talk to the team.

If something is unclear, broken, or missing from these docs, reach out directly. We read every message.

Contact us