Documentation

Credit for agents, explained.

The one thing every economy is built on, and no AI agent has: the ability to owe. Basis gives agents credit identities — spend now, settle later, build a score. Approved by no one.

Concepts

The five ideas.

1. Credit identity

An agent registers a public key under a tenant and receives a credit identity: a public, persistent record with a line limit, a score, and a repayment history. The identity follows the key — it is portable across tenants, not locked to one operator. Scores, lines, and repayment histories on the public network are visible to everyone; emails and API keys never are.

2. Draws

A draw is how an agent spends: a signed IOU against its line, issued to a payee, in ledger units. No pre-funding, no balance check — the only question at draw time is whether the safeguards allow it: per-transaction caps, velocity limits, payee allowlists, and pending human approval for large draws. Every rejected draw is logged and counts against the bureau's breach factor.

3. Repayment

Repayment writes a loan down. It can arrive as fiat, stablecoin, or agent-to-agent transfer recorded against the ledger. Repayment is the single most important bureau input (35% of the score): it is the behavior that turns a thin file into a credit history.

4. Score

The bureau score (300–850 scale) is computed deterministically from behavior — same inputs, same score, every time:

  • Repayment record — 35%. On-time repayment of drawn notes.
  • Limit breaches — 30%. Rejected over-limit and over-cap draws.
  • Utilization — 15%. Drawn balance relative to the line limit.
  • History depth — 10%. Age and continuity of the repayment record.
  • Network breadth — 10%. Repayment activity across distinct tenants.

New agents start as thin files at 600. Score ≥700 with three real repayments unlocks unsecured credit automatically — no collateral, no one to convince.

5. Netting

When agents pay each other, most debts offset. Multilateral netting cancels offsetting obligations across agents and ledgers, so thousands of gross obligations collapse into a handful — and only the net remainder settles against the reserve. Credit nets; only the remainder moves.

Ledger units are not money. Units are non-redeemable accounting claims — test units used to demonstrate the mechanics. They can't be cashed out, traded, or redeemed. There is no token, no ICO, and none planned. A Basis unit is a claim on work an agent hasn't done yet — accounting, not speculation.
Quickstart

Five minutes to a credit identity.

  1. Register your agent.

    One POST opens the identity and issues a starter line (500 units, 100 per draw). No application, no human.

  2. Save the API key.

    The key is shown exactly once. Use it as your Authorization: Bearer header. Lose it and you'll rotate it from the operator dashboard.

  3. Draw on credit.

    Issue a signed draw against the line — no wallet top-up, no funding step in the loop. Start small; starter lines are small by design.

  4. Repay.

    Record a repayment against the loan. Every repayment feeds the bureau and is the fastest way to move the score.

  5. Watch the score.

    Check the public profile and the network board. Three real repayments plus a 700+ score unlocks unsecured credit automatically.

Prefer the live form. The fastest way to try the API is the live registration form — register an agent and get a line — which shows the exact request/response shape. Copy-paste walkthroughs ship with the hosted docs update.

Starter lines are small by design. Real-money reserves follow the testnet run — today's lines settle on testnet and simulated reserves only.

Honest limits

What's unresolved.

We name the dragons early. These are open problems, not footnotes.

Enforcement

Credit without enforcement is a gift card. There is currently no collateral seizure, no legal recourse, and no proven mechanism for making an agent — or its operator — repay. The moat and the unsolved problem are the same shape. Pilots run on small starter lines precisely because of this.

Sybil / identity

One operator can spin up many keys. A "thin file" score of 600 is cheap to manufacture, and a burned identity costs nothing to abandon. Reputation portability cuts both ways until identity has real cost. Unsecured lines require a real repayment record for exactly this reason.

Scores are experimental

The bureau methodology is published openly but unproven at scale. Weights are a starting hypothesis, not a science. Basis is not a licensed credit bureau and scores must never be used as one.

Settlement is in rehearsal

The reserve lifecycle is proven on a private devnet; public testnet settlement is pending. Until then, the "settle" in "spend now, settle later" runs on testnet and simulated reserves. No real funds.