DocsStart free
Docs menu · Overview

Neltava documentation

Neltava decides whether an AI agent should spend, before money moves: against the rules you set and the purpose you gave it. Start here.

Updated

What Neltava does

Your agents can spend: they buy data, top up API credits, book ads, renew tools. Neltava decides whether they should — right before the money moves. Every spend is checked against the authority you delegated to the agent (limits, budgets, merchants, categories, pace) and against its purpose (what the spending is for). The answer comes back in milliseconds, with a stable reason code and a human-readable explanation, and is recorded in a tamper-evident history.

Neltava never holds or moves money. Your agent keeps paying the way it already does.

The flow

one spend
agent ──"may I spend 49 USD at Statista, for this?"──▶ Neltava
                                                        │ rules → budget → purpose
agent ◀──────────── ALLOW · LIMIT · REVIEW · BLOCK ─────┘
  │
  └─ pays the way it already does (or, in Enforce, through your payment adapter)

Four answers

VerdictMeaning
ALLOWWithin the agent’s authority and purpose. Spend it.
LIMITAllowed, but for less: spend at most the authorized amount. Only when the request said it can be capped.
REVIEWA person decides — over a threshold, outside a limit, or doubtful for the purpose.
BLOCKOutside the delegated authority. Only deterministic rules block; an AI assessment never does.

How each verdict is reached →

Shadow and Enforce

Every agent starts in Shadow Mode: Neltava decides and records every spend, but the answer your agent acts on is always to proceed. You see what Neltava would have stopped or sent to review, without blocking a single action. When the Shadow report shows the evidence is strong enough, you switch that agent to Enforce, and the verdict takes effect — at the payment point, with single-use capabilities.

Where to start

Quickstart →

From sign-up to your first decision in about ten minutes.

MCP server →

Claude Code, Cursor, Claude Desktop — add a server, no code.

TypeScript SDK →

npm install neltava, one call before your agent pays.

HTTP API →

One HTTP request from any language.

The basics

  • API: https://api.neltava.com. Authenticate with an API key in x-api-key (MCP uses Authorization: Bearer).
  • Keys: each agent gets its own key, which can only ask for decisions as that agent. Admin, reviewer, read-only and payment-adapter keys are separate. Keys and scopes →
  • Money is integer minor units in the API (amount_minor: 4900 is 49.00 USD). The SDK and MCP take major units.
  • Retries are safe: every decision request carries an idempotency-key; repeating it returns the same decision.
  • Console: console.neltava.com — agents, authority, decisions, reviews, the Shadow report, keys and webhooks.

Next

Quickstart →