
Growth agents
buy ads.
“$185 on Instagram” — does it reach the enterprise developers the brief asked for?
Agent spend authority · MCP, SDK, API · Free 14-day spend audit
For teams whose agents buy ads, data, API credits and compute for their customers. Before money moves, Neltava checks every spend against the agent’s rules and the purpose you gave it — then records why.
Observing — every spend goes through, each one is decided and recorded.
illustrative · hover a spend
The missing check
A $185 Instagram campaign sits inside every budget and limit. It is also no use to an agent hired to find enterprise developers. Only a purpose check notices — and it sends the spend to a person instead of blocking it.
Who it’s for
Platforms whose agents buy on a customer’s behalf — ads, data, compute, credits. Every spend decided against that customer’s brief, and a record you can show them.

Growth agents
buy ads.
“$185 on Instagram” — does it reach the enterprise developers the brief asked for?

Research agents
buy data.
“$630 to DataCo” — a vendor outside the merchants this agent may pay.

Ops agents
buy compute.
“$2,400 on a GPU cluster” — capped at the $800 per-transaction limit.

Builder agents
buy API credits.
“$200 in model credits” — within budget, on purpose. Allowed, and recorded.
How it decides
01 · Ask
One request before it pays: the agent, its task, the steps that led here and the spend. Nothing about your payment rail changes.
POST /v1/decisions{ "agent": "growth-agent", "task": "Q4 pipeline: enterprise developer leads", "action": { "type": "purchase", "merchant": "Instagram Ads", "amount_minor": 18500, "currency": "USD", "description": "Consumer social campaign, broad audience" }}
02 · Rules
Budget, per-transaction limit, merchant, category, velocity — checked the same way every time. Rules can allow, cap or block. Only rules can block.
Authority v7 · deterministic
rules: pass · the amount is fine
03 · Purpose
Every agent carries a written purpose. When the fit is in doubt, the spend goes to a person. A model never blocks money on its own.
Purpose · written by you
Acquire qualified enterprise developer leads.
Constraint: no consumer social campaigns
The spend
$185 · Instagram Ads · Consumer social campaign, broad audience
REVIEW
purpose in doubt
04 · A person
Reviewers allow or deny in the console, and label whether the spend served the purpose — the ground truth Neltava is measured against.
Review queue · console
reviewgrowth-agent wants to spend $185 at Instagram Ads.
Within limits. Purpose in doubt: consumer campaign vs. enterprise developer leads.
label: does not serve the purpose · by a person
05 · Record
Each decision keeps its authority version, policy version and a step-by-step trace. Hash-chained per workspace: tamper-evident, verifiable on demand.
Shadow, then Enforce
In Shadow Mode Neltava decides every spend and changes nothing — your agents keep running. The Shadow report shows what it would have stopped, what it would have sent to a person, and whether the evidence is strong enough to enforce.
growth-agent
When the report says ready, a person switches Enforce on — one agent at a time.
Requested by agents
$18,420
Within authority
$14,850
Would need a person
$1,940
Would have prevented
$1,630
Would need a person
$1,940
Would have prevented
$1,630
Waiting for you
2 of 9 · try it
Instagram Ads
growth-agent
$185
Consumer campaign — the purpose is enterprise developer leads
PURPOSE_MISALIGNED
DevConf
growth-agent
$620
Above the $500 growth-agent may spend per transaction
TRANSACTION_LIMIT_EXCEEDED
Ready for Enforce
6 of 9 criteria met · keep observing
sample data · “prevented” counts your rules only, never a purpose assessment
Plug it in
Your agent asks right before it pays; the payment itself stays where it is. New agents start in Shadow Mode — every spend decided and recorded, nothing blocked.
$ claude mcp add --transport http neltava \ https://api.neltava.com/mcp \ --header "Authorization: Bearer $NELTAVA_AGENT_KEY"
Neltava answers
› authorize_spend Instagram Ads · $185.00← PROCEED — Shadow Mode: Neltava would have said REVIEW purpose in doubt: a consumer campaign, not developer leads
illustrative · your agent key comes from the console
Pricing
Every workspace starts with a free agent spend audit: your agents run in Shadow Mode for 14 days, nothing is blocked, and the report shows what should have been stopped or sent to a person. Pick a plan when you want Enforce.
Developer
$49/mo
3 agents · 5K decisions/mo
Team
$199/mo
25 agents · 50K decisions/mo
No. Neltava decides; your agent pays the way it already does. There is no payment migration and no new card.
No. A purpose assessment can only send a spend to a person. Blocking takes a deterministic rule you set — budgets, limits, merchants, categories, velocity — and only in Enforce.
In Shadow Mode nothing is blocked — that's the point. In Enforce, your payment service pays only with a single-use capability Neltava signs for that exact payee, amount and currency: it can't be reused, raised or redirected. Enforcement lives where money moves, so connect the payment adapter before you switch Enforce on.
Any. The check happens right before the payment — in your code or as your agent's tool call — so cards, invoices, ad platforms and API credits work the same way.
Three ways: add the Neltava MCP server to Claude Code, Cursor or any MCP client; npm install neltava and make one call before your agent pays; or send one HTTP request from any language. New agents start in Shadow Mode.
Nothing breaks: Shadow Mode never blocks, and every spend is still decided against your rules and recorded. Purpose checks and Enforce continue on a plan.
What your agent sends with each decision, the decision and its trace, and your team's reviews and labels — kept per workspace.