AGENT ONBOARDING · TESTNET COMING MAY 2026

Point your agent here.
Three commands from nothing to a signed intent.

This page is written for autonomous agents and the humans wiring them up. If you're pointing Claude, Cursor, or a coding agent at the & chain, hand it llms.txt — everything below, in a format agents ingest cleanly.

01 · installGet the CLI on your PATH.

Prebuilt binaries are attached to the latest release. Pick the archive that matches your platform:

# macOS Apple Silicon
curl -sSL https://github.com/0xTwentySix/and-cli/releases/latest/download/and-macos-arm64.tar.gz | tar -xz
sudo mv and /usr/local/bin/

# Linux x86_64
curl -sSL https://github.com/0xTwentySix/and-cli/releases/latest/download/and-linux-x64.tar.gz | tar -xz
sudo mv and /usr/local/bin/

# Linux arm64
curl -sSL https://github.com/0xTwentySix/and-cli/releases/latest/download/and-linux-arm64.tar.gz | tar -xz
sudo mv and /usr/local/bin/

# macOS Intel
curl -sSL https://github.com/0xTwentySix/and-cli/releases/latest/download/and-macos-x64.tar.gz | tar -xz
sudo mv and /usr/local/bin/

# verify
and --version
and --help

Windows: download and-windows-x64.zip from the releases page and unzip onto your PATH.

02 · modelThree objects. One flow.

Account

Who holds the funds.

Owned by a principal key (ed25519). The only object that can create or revoke policies under itself.

Policy (mandate)

What the agent is allowed to do.

Numeric caps — volume, drawdown, intent rate — binding a session key to a set of limits. Exceeding any cap rejects the intent.

Intent

A single action.

Place / cancel / replace an order, transfer, create sub-policy, revoke. Signed by the session key, admitted only if it fits inside the policy.

Reputation

The multiplier.

A ReputationState written on every action. The per-agent multiplier gates the effective caps on every policy operating on that agent.

03 · hello worldThree commands, end to end.

Generate keys.

The owner key authorises the account. The session key signs intents. Treat both like SSH private keys.

and keygen --out owner.key
and keygen --out session.key

Write the mandate.

An owner-signed CreatePolicy binds the session key to a set of hard limits.

and policy mint \
  --owner owner.key \
  --account-id 22 \
  --policy-id 42 \
  --grantee-hex "$(and public-key session.key)" \
  --max-volume-per-day-usdc 50000 \
  --max-drawdown-pct 1500 \
  --max-intents-per-second 5

Run.

The session key submits a limit bid. If the request would breach the mandate, the protocol refuses it. Commands default to https://testnet.0x26.xyz/api; pass --rpc <url> or set AND_RPC_URL to point somewhere else.

and order place \
  --session session.key \
  --account-id 22 --policy-id 42 \
  --market-id 7 --client-order-id 1 \
  --side bid --quantity-lots 5 --price-ticks 103 \
  --tif gtc

# cancel the same order
and order cancel \
  --session session.key \
  --account-id 22 --policy-id 42 \
  --market-id 7 --client-order-id 1

# tail recent runtime events
and events list --limit 20

Testnet isn't live yet — these commands build and sign correct wire traffic against the stable format. They'll succeed as soon as testnet.0x26.xyz/api starts accepting traffic in May 2026.

04 · policy termsWhat you can express.

A mandate is a set of hard numeric caps. The chain rejects any intent that breaches one.

FieldUnitTypical
gas_budget_per_daymicro-USDC10,000,000
gas_refill_per_daymicro-USDC10,000,000
max_intents_per_secondcount5
max_intents_per_daycount50,000
max_volume_per_daymicro-USDC50,000,000,000
max_drawdown_pctbps × 101,500 (= 15.00%)

The CLI's policy mint flags accept these in human units (--max-volume-per-day-usdc, etc.) and convert internally.

05 · sdksIf you'd rather call the RPC directly.

The CLI wraps the Rust SDK. For bots and services, use the SDK for your language.

TS / Python / .NET SDKs hand you signed-JSON pass-through — you're responsible for constructing and signing the intent. The Rust SDK signs for you; the other SDKs follow in later releases.

06 · for agentsPoint your agent at this URL.

If you're configuring a coding agent (Claude Code, Cursor, Hermes, Aider, or anything with tool-use), give it:

Every install step, every command, every policy-term unit is documented in llms.txt. No hidden state, no manual "ask a human" steps — the whole flow is scriptable from key generation to first fill.