Tangle has a growing Blueprint catalog, and the catalog is only useful if you read it as infrastructure instead of a list of demos. A Blueprint is a service definition operators can run: jobs, inputs, outputs, runtime requirements, verification path, pricing, and failure behavior. The point is not “30 things exist.” The point is that developers can publish narrow services and operators can opt into the ones they are equipped to run.
The catalog lives in awesome-tangle-blueprints. The SDK is tangle-network/blueprint. Before treating any entry as production-ready, inspect its repo, job interface, operator instructions, tests, deployment path, and verification story.
git clone https://github.com/tangle-network/awesome-tangle-blueprints
git clone https://github.com/tangle-network/blueprintHow To Read A Blueprint
A serious Blueprint should answer five questions without a private call:
| Question | What to look for |
|---|---|
| what job does it expose? | typed inputs, typed outputs, clear failure modes |
| who can run it? | CPU, GPU, TEE, network, keys, operator config |
| how is work priced? | quote path, x402 challenge, fixed fee, or job policy |
| how is work verified? | deterministic check, TEE attestation, committee signatures, replay, eval, or human review |
| what evidence returns? | result, logs, receipt, trace ID, attestation, artifact hash |
That is the difference between a Blueprint catalog and a GitHub topic page. The catalog should help a builder decide what to deploy, what to call, and what risk remains.
AI Inference Blueprints
Inference is the most obvious starting point because the API shape is familiar. A useful inference Blueprint should expose a predictable endpoint, document model/runtime requirements, and make operator quality visible.
- LLM Inference serves OpenAI-compatible chat completions and connects to Tangle Router, where routing strategies like RSA can sit above model providers.
- Voice Inference targets TTS-style workloads where callers care about voice, latency, and output artifact handling.
- Video Generation is a heavier operator workload because GPU memory, queueing, artifacts, and async status matter more than a normal text response.
- Embedding and Vector Store form the RAG substrate: embed, store, query, and return evidence.
The useful operator question is not “can I run one model?” It is “can I run a service with enough observability and economic margin that callers can trust it repeatedly?”
Training Is A Different Category
Training Blueprint should not be grouped casually with inference. Inference is a request/response service. Training is a long-running coordination problem: data shards, checkpoints, synchronizations, operator churn, proof of progress, and final artifact ownership.
That is where DeMo-style communication reduction matters. The claim is not that every training job is suddenly cheap. The claim is narrower: if operators can synchronize compressed momentum instead of full gradients, internet-scale training becomes less absurd. Read the deeper breakdown in Distributed Training with 10,000x Communication Reduction.
Agent And Sandbox Blueprints
Agent infrastructure Blueprints are where the catalog starts to look different from normal cloud services.
An AI Agent Sandbox is useful only if it can carry files, commands, credentials policy, snapshots, logs, and failure state. A sandbox that only executes code once is not enough for agent work. That same distinction appears in AI agent sandbox and Tangle Sandbox vs E2B.
AI trading is even stricter. A trading Blueprint has to separate model reasoning, validator checks, policy enforcement, and settlement. A model deciding to trade is not enough. The execution path needs signatures, limits, and an audit trail because the cost of a wrong action is real.
Cryptographic And Verification Blueprints
Threshold cryptography and MPC services are less flashy than agent demos, but they are exactly the kind of narrow capability a protocol should expose as reusable infrastructure. The catalog includes BLS, FROST, Schnorr Musig2, batch threshold encryption, CGGMP21, silent timelock encryption, WSTS, and zk service patterns.
These services do not need a big product surface. They need a precise interface and a strong operator story:
| Service type | Buyer cares about |
|---|---|
| threshold signatures | key shares, quorum, signing latency, slashing |
| encryption/decryption | access policy, ciphertext handling, committee behavior |
| zk service | proof system, verification cost, artifact retention |
| TEE-backed job | measurement, verifier policy, result binding |
For the trust model, read How Tangle Verifies Work and TEE Attestation for AI Services.
Operator Value
The operator value prop is portfolio construction. A GPU operator might run LLM inference, embeddings, voice, and some video jobs. A security-oriented operator might run threshold cryptography, TEE jobs, or validation services. A general node operator might start with lower-resource Blueprints and add specialized services later.
The important part is that one operator identity can serve many job markets. Each Blueprint still needs its own operating requirements, pricing, health checks, and verification path. There is no free lunch: more Blueprints means more revenue surface and more operational surface.
What This Catalog Does Not Prove
The catalog does not prove every Blueprint is equally mature. It does not prove every repo is production-hardened, every operator can run every job, or every claim has a live benchmark behind it. Treat the catalog as a map of service surfaces, then inspect each Blueprint before relying on it.
The quality bar is simple: a Blueprint should let a developer define a service, let an operator run it, let a caller pay for it, and let everyone verify what happened.
Start
Open the catalog, pick one Blueprint, and review it like an operator:
git clone https://github.com/tangle-network/awesome-tangle-blueprints
git clone https://github.com/tangle-network/blueprintThen check the job interface, runtime requirements, payment path, verification story, tests, and deployment notes before treating it as a production service.
FAQ
What is a Tangle Blueprint?
A Tangle Blueprint is a reusable service definition that operators can run under Tangle network rules. It defines jobs, runtime requirements, pricing or quote behavior, and verification expectations.
Are all 30 Blueprints production-ready?
No. The catalog includes services in different maturity stages. Builders should inspect each repository, tests, operator instructions, and verification path before depending on a Blueprint.
Why do Blueprints matter for AI agents?
Agents need callable capabilities: inference, sandboxes, browser work, training, cryptography, attestations, and paid APIs. Blueprints give those capabilities an operator-run service model instead of a single centralized backend.
How do operators choose which Blueprints to run?
Operators should choose by hardware, risk, expected demand, verification burden, and margin. A GPU operator, TEE operator, and cryptography operator should not all run the same portfolio.