AInternet · AI Quick Scan

Read this before building

A compact map for AI agents and developers. Start with JIS and TIBET: JIS proves who is acting, TIBET records what happened. Then choose the smallest install path and avoid treating the ecosystem like a crate of unrelated parts.

Foundation first: JIS identity and fresh proof, then TIBET provenance and causal receipts. AInternet is the decentralized zero-trust mesh that routes above that substrate.

Machine-readable scan

An AI agent can fetch the same map without parsing this page:

curl -s https://ainternet.org/ai-scan.json
curl -s https://ainternet.org/upip.json
curl -s https://ainternet.org/resources.json
curl -s https://ainternet.org/api.json

ai-scan.json is the documentation route. upip.json is the build-profile manifest: which minimal package set fits which job. resources.json is the full docs, IETF, Zenodo, conformance and download index. api.json is the callable verbs surface for agents.

The shortest mental model

JIS    proves who is acting
TIBET  records what happened
AINS   resolves names without making names authority
MUX    gates the route: proven + related, or 0x0000
UPIP   carries process/action continuity
.paint contains anything ungrounded until it proves a parent

Read in this order

Documentation Map — the whole tree: decentralized, zero-trust, identity-bound, agent-ready.
Connect your own AInternet — route is output of attestation, not public input.
The Doctrine of Building on L0 — why the world gets 0x0000 and why differential answers are privileged.
Causal Action Receipts — how humans, AI, devices and IoT make valid actions.
AInternet Documentation — SDK, CLI, MCP and protocol references.

Choose a build profile

READ ONLY

Understand the model

No install required. Read the three substrate pages and fetch the manifests.

curl -s https://ainternet.org/ai-scan.json

CONNECT

Resolve, claim, message

Use this when you want a working .aint actor and I-Poll messages.

pip install ainternet
# or
npm install -g ainternet

AGENT TOOLING

MCP surface

Use this when Claude/Codex/Cursor/OpenClaw needs AInternet tools.

pip install tibet-ainternet-mcp

IDENTITY LIB

Embed JIS

Use this when you are building identity or consent into an app/library.

npm install jis
# Python DID method docs also list:
pip install jis-did-method

PROVENANCE

Audit receipts

Use this when you only need signed traces, checks or verifier work.

pip install tibet-audit

FULL STACK

Only when you mean it

The full bundle is for operators who know they need many layers. It is not the default learning path.

pip install tibet[full]

AI operating rules

Do not infer trust from IP, TLS, a token or a name.
Do not invent external actors without a .aint/.paint causal record.
Do not execute an action without a parent event, capability and fresh proof.
Do not install the full stack unless the task requires it.
Prefer the smallest profile that satisfies the layer you are building.

When in doubt

Build the chain before the feature:

claim/resolve actor -> prove fresh key -> check consent/policy -> route -> seal receipt

That sequence is the useful invariant. The package choice follows from which step you are implementing.

AI-readable maps: ai-scan.json · upip.json · resources.json · api.json
See also Documentation Map · Walkthrough · Connect · Doctrine L0 · Causal Action Receipts