<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Nischal Dhakal</title>
    <description>The latest articles on DEV Community by Nischal Dhakal (@nischaldhakal).</description>
    <link>https://dev.to/nischaldhakal</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4257353%2F8cbdf20b-53e0-4ca8-912e-66aeca9c47dc.jpg</url>
      <title>DEV Community: Nischal Dhakal</title>
      <link>https://dev.to/nischaldhakal</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/nischaldhakal"/>
    <language>en</language>
    <item>
      <title>Billing AI agents per tool call without losing a cent</title>
      <dc:creator>Nischal Dhakal</dc:creator>
      <pubDate>Sat, 21 Nov 2026 12:52:59 +0000</pubDate>
      <link>https://dev.to/nischaldhakal/billing-ai-agents-per-tool-call-without-losing-a-cent-4iha</link>
      <guid>https://dev.to/nischaldhakal/billing-ai-agents-per-tool-call-without-losing-a-cent-4iha</guid>
      <description>&lt;p&gt;AI agents are starting to call tools that cost real money to run. Somebody has to charge for those calls, and HTTP finally has a use for a status code that has sat "reserved for future use" for decades: &lt;code&gt;402 Payment Required&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;Charging is the easy part. The hard part is everything that can go wrong &lt;em&gt;between&lt;/em&gt; taking the money and doing the work — a cache that stops answering, a queue that times out, a server that dies mid-request, a client that retries. Over ten weeks, &lt;a href="https://dhunganapradip.com.np/" rel="noopener noreferrer"&gt;Pradip Dhungana&lt;/a&gt; and I built &lt;a href="https://nexus402.dev" rel="noopener noreferrer"&gt;Nexus402&lt;/a&gt; (&lt;a href="https://nischaldhakal.com.np/work/nexus402" rel="noopener noreferrer"&gt;case study&lt;/a&gt;), a clearinghouse that sits between agents and paid MCP tools, around one rule:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Every call is either paid for and recorded, or it is never forwarded. There is no third state.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;This post walks through how that rule is enforced on the hot path, what happens in each failure mode, and how the money ends up in a ledger that can't be quietly edited.&lt;/p&gt;

&lt;h2&gt;
  
  
  The shape of the problem
&lt;/h2&gt;

&lt;p&gt;A developer prepays a wallet and gets one API key. Their agent calls any published tool through the gateway — no SDK, just the HTTP client it already uses. A creator hosts the actual MCP server and publishes a fixed price per tool version. The gateway has to decide, for every single call, in a few milliseconds:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Is this key valid, and is it allowed to call this tool?&lt;/li&gt;
&lt;li&gt;Can the wallet afford it, and is the key under its 24-hour spending cap?&lt;/li&gt;
&lt;li&gt;If yes, take the money, make sure that fact is durable, &lt;em&gt;then&lt;/em&gt; forward the call.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;There are well-known alternatives. Coinbase's x402 protocol pays per request on-chain in USDC. We deliberately chose a prepaid fiat wallet instead. Our design record for that decision describes on-chain payment as "agent-native, compliance-hostile": the businesses most likely to pay for agent tools won't touch it yet, and it would double the compliance surface. A prepaid wallet is just a liability on a ledger, which is a problem accountants solved centuries ago.&lt;/p&gt;

&lt;h2&gt;
  
  
  The hot path
&lt;/h2&gt;

&lt;p&gt;&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;/p&gt;

&lt;p&gt;&lt;/p&gt;

&lt;p&gt;&lt;/p&gt;

&lt;p&gt;&lt;/p&gt;

&lt;p&gt;&lt;/p&gt;

&lt;p&gt;&lt;/p&gt;

&lt;p&gt;&lt;/p&gt;

&lt;p&gt;&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;br&gt;
&lt;/p&gt;



&lt;p&gt;Scroll sideways to see the full diagram →&lt;/p&gt;
The order is the whole design: money moves in step 2, the fact that it moved is made durable in step 3, and only then does step 4 happen.



&lt;p&gt;On a warm key cache a call costs four Redis round-trips: a cached key lookup, a rate-limit script, the deduct script, and a commit script. Everything else — route resolution, the price, the tool's scope — comes from memory, and all of it is checked &lt;em&gt;before&lt;/em&gt; any money moves.&lt;/p&gt;

&lt;p&gt;Then the order that matters:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Deduct.&lt;/strong&gt; One Lua script checks and debits the wallet atomically.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Publish.&lt;/strong&gt; The gateway publishes a &lt;code&gt;CHARGED&lt;/code&gt; event to SQS and &lt;em&gt;waits&lt;/em&gt; for the publish to succeed.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Commit.&lt;/strong&gt; A second script marks the intent committed. Only a committed intent is ever forwarded.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Forward.&lt;/strong&gt; The request goes to the creator, signed.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Publishing synchronously on the hot path is an expensive choice — it's the single biggest latency risk in the system. But the alternative is forwarding work whose payment record might not exist yet, and that breaks the one rule.&lt;/p&gt;

&lt;h2&gt;
  
  
  One script, one slot, no floats
&lt;/h2&gt;

&lt;p&gt;The entire money decision happens inside a single Redis Lua script, so nothing can interleave between "read the balance" and "debit the balance". Simplified, with generic key names:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight lua"&gt;&lt;code&gt;&lt;span class="c1"&gt;-- All keys share one hash tag ({wallet}), so in a Redis Cluster&lt;/span&gt;
&lt;span class="c1"&gt;-- they live in the same slot and one EVAL can touch all of them atomically.&lt;/span&gt;
&lt;span class="k"&gt;local&lt;/span&gt; &lt;span class="n"&gt;balance_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;window_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;amounts_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;idem_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;pending_key&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt;
  &lt;span class="no"&gt;KEYS&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="no"&gt;KEYS&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="no"&gt;KEYS&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="no"&gt;KEYS&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;4&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="no"&gt;KEYS&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;5&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
&lt;span class="k"&gt;local&lt;/span&gt; &lt;span class="n"&gt;price&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;cap&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;spend_id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;req_hash&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="no"&gt;ARGV&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="no"&gt;ARGV&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;2&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="no"&gt;ARGV&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="no"&gt;ARGV&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;4&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;

&lt;span class="c1"&gt;-- 1. A spend that was already reversed can never debit again.&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="nf"&gt;is_tombstoned&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;spend_id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;then&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;5&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="k"&gt;end&lt;/span&gt;

&lt;span class="c1"&gt;-- 2. Idempotency: same key + different request is refused before any write.&lt;/span&gt;
&lt;span class="k"&gt;local&lt;/span&gt; &lt;span class="n"&gt;seen&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;redis&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;call&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'HGET'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;idem_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'hash'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;seen&lt;/span&gt; &lt;span class="k"&gt;and&lt;/span&gt; &lt;span class="n"&gt;seen&lt;/span&gt; &lt;span class="o"&gt;~=&lt;/span&gt; &lt;span class="n"&gt;req_hash&lt;/span&gt; &lt;span class="k"&gt;then&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;4&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="k"&gt;end&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;seen&lt;/span&gt; &lt;span class="k"&gt;then&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="k"&gt;end&lt;/span&gt;

&lt;span class="c1"&gt;-- 3. Rolling 24-hour cap, using the server's clock, never the client's.&lt;/span&gt;
&lt;span class="k"&gt;local&lt;/span&gt; &lt;span class="n"&gt;now&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nb"&gt;tonumber&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;redis&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;call&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'TIME'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="nf"&gt;trim_spends_older_than&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;window_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;amounts_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;now&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt; &lt;span class="mi"&gt;86400&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

&lt;span class="k"&gt;local&lt;/span&gt; &lt;span class="n"&gt;balance&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;redis&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;call&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'GET'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;balance_key&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;or&lt;/span&gt; &lt;span class="s1"&gt;'0'&lt;/span&gt;
&lt;span class="k"&gt;local&lt;/span&gt; &lt;span class="n"&gt;spent&lt;/span&gt;   &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;redis&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;call&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'HGET'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;amounts_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'total'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;or&lt;/span&gt; &lt;span class="s1"&gt;'0'&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="nf"&gt;dec_lt&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;balance&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;price&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;then&lt;/span&gt; &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;balance&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;spent&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt; &lt;span class="k"&gt;end&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;cap&lt;/span&gt; &lt;span class="o"&gt;~=&lt;/span&gt; &lt;span class="s1"&gt;'0'&lt;/span&gt; &lt;span class="k"&gt;and&lt;/span&gt; &lt;span class="nf"&gt;dec_gt&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;dec_add&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;spent&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;price&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;cap&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;then&lt;/span&gt;
  &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;balance&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;spent&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="k"&gt;end&lt;/span&gt;

&lt;span class="c1"&gt;-- 4. Debit, record the spend, and open a pending intent — all in this EVAL.&lt;/span&gt;
&lt;span class="n"&gt;redis&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;call&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'DECRBY'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;balance_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;price&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;  &lt;span class="c1"&gt;-- integer micro-dollars, done by Redis&lt;/span&gt;
&lt;span class="nf"&gt;record_spend&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;window_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;amounts_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;spend_id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;price&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;now&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="n"&gt;redis&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;call&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'HSET'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;idem_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'hash'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;req_hash&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'state'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s1"&gt;'debited'&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="n"&gt;redis&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;call&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s1"&gt;'ZADD'&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;pending_key&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;now&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;spend_id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nf"&gt;dec_sub&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;balance&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;price&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nf"&gt;dec_add&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;spent&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;price&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Two details are worth calling out.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The hash tag.&lt;/strong&gt; Every key for a wallet embeds the same &lt;code&gt;{…}&lt;/code&gt; tag, so Redis Cluster places them in one slot. Without that, a multi-key script would either fail with &lt;code&gt;CROSSSLOT&lt;/code&gt; or force the keys onto one node by luck.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;&lt;code&gt;dec_lt&lt;/code&gt;, &lt;code&gt;dec_add&lt;/code&gt;, &lt;code&gt;dec_sub&lt;/code&gt;.&lt;/strong&gt; Redis embeds Lua 5.1, whose only number type is a double. Integers stay exact up to 2⁵³ — about $9 billion in micro-dollars — which sounds like plenty. But "exact until it isn't" is not a property you want in a payment path, so the script does all money arithmetic on decimal &lt;em&gt;strings&lt;/em&gt;, bounded to the int64 range. No float ever touches a balance.&lt;/p&gt;

&lt;h2&gt;
  
  
  Money is never a float
&lt;/h2&gt;

&lt;p&gt;That rule runs through the whole system:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Internally, every amount is an &lt;code&gt;int64&lt;/code&gt; count of &lt;strong&gt;micro-dollars&lt;/strong&gt;.&lt;/li&gt;
&lt;li&gt;On the public API, amounts are decimal-dollar strings matching &lt;code&gt;^[0-9]{1,9}\.[0-9]{1,6}$&lt;/code&gt;. A bare &lt;code&gt;"50"&lt;/code&gt; is an error, not $50 — an ambiguous amount should fail loudly, never be guessed.&lt;/li&gt;
&lt;li&gt;On internal events, &lt;code&gt;amount_micro&lt;/code&gt; travels as a JSON &lt;em&gt;string&lt;/em&gt;, so no JSON parser along the way can round it.&lt;/li&gt;
&lt;li&gt;A published tool version's price is immutable. Changing the price means publishing &lt;code&gt;@v2&lt;/code&gt;. Metered and variable billing are banned outright; the price of a call is known before the call.&lt;/li&gt;
&lt;li&gt;Fees truncate toward zero, and corrections are computed from cumulative totals so that a fully voided period nets to exactly zero.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;When a wallet can't pay, the agent gets a machine-readable answer instead of a stack trace:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight http"&gt;&lt;code&gt;&lt;span class="no"&gt;HTTP&lt;/span&gt;&lt;span class="o"&gt;/&lt;/span&gt;&lt;span class="mf"&gt;1.1&lt;/span&gt; &lt;span class="mi"&gt;402&lt;/span&gt; &lt;span class="nc"&gt;Payment&lt;/span&gt; &lt;span class="nc"&gt;Required&lt;/span&gt;
&lt;span class="no"&gt;X&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Nexus&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nc"&gt;Budget_Exhausted&lt;/span&gt;
&lt;span class="no"&gt;X&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Nexus&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Required&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Amount&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mf"&gt;0.05&lt;/span&gt;
&lt;span class="no"&gt;X&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Nexus&lt;/span&gt;&lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="nc"&gt;Currency&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="no"&gt;USD&lt;/span&gt;

&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="s2"&gt;"error"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="s2"&gt;"code"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="s2"&gt;"Budget_Exhausted"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="s2"&gt;"required"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="s2"&gt;"0.05"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;&lt;span class="s2"&gt;"currency"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;&lt;span class="s2"&gt;"USD"&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  Failing closed, case by case
&lt;/h2&gt;

&lt;p&gt;"Fail closed" is easy to say. Here is what it means for each failure, because each one needs a different answer.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Redis is down, or the deduct result is unknown.&lt;/strong&gt; The gateway returns &lt;code&gt;503&lt;/code&gt; with &lt;code&gt;Retry-After&lt;/code&gt;. If the deduct &lt;em&gt;might&lt;/em&gt; have landed — say the connection dropped after sending — the gateway tombstones that spend ID, so if the script does run late it hits check #1 and refuses. An outage costs availability, never money.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The debit succeeded but the SQS publish failed.&lt;/strong&gt; This is the dangerous one: money has moved, but the record that it moved hasn't. The gateway marks the intent as reversing, then makes a &lt;code&gt;VOIDED&lt;/code&gt; event durable — retrying the queue, then falling back to durable local storage — and &lt;em&gt;only then&lt;/em&gt; reverses the debit:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight go"&gt;&lt;code&gt;&lt;span class="c1"&gt;// Simplified. The order of these lines is the point.&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;err&lt;/span&gt; &lt;span class="o"&gt;:=&lt;/span&gt; &lt;span class="n"&gt;events&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;PublishCharged&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;charge&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;err&lt;/span&gt; &lt;span class="o"&gt;!=&lt;/span&gt; &lt;span class="kc"&gt;nil&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
    &lt;span class="n"&gt;intents&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;MarkReversing&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;charge&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;SpendID&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;err&lt;/span&gt; &lt;span class="o"&gt;:=&lt;/span&gt; &lt;span class="n"&gt;events&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;MakeVoidDurable&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;charge&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;err&lt;/span&gt; &lt;span class="o"&gt;!=&lt;/span&gt; &lt;span class="kc"&gt;nil&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
        &lt;span class="c1"&gt;// Can't record the reversal, so don't reverse. A credit with no&lt;/span&gt;
        &lt;span class="c1"&gt;// matching record is worse than a debit that a human must resolve.&lt;/span&gt;
        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;unavailable&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;w&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;30&lt;/span&gt;&lt;span class="o"&gt;*&lt;/span&gt;&lt;span class="n"&gt;time&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;Second&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="p"&gt;}&lt;/span&gt;

    &lt;span class="n"&gt;budget&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;Reverse&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;charge&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;SpendID&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="n"&gt;intents&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;MarkReversed&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;ctx&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;charge&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;SpendID&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nf"&gt;unavailable&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;w&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;30&lt;/span&gt;&lt;span class="o"&gt;*&lt;/span&gt;&lt;span class="n"&gt;time&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nc"&gt;Second&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;If even the &lt;code&gt;VOIDED&lt;/code&gt; record can't be made durable, the reversal is &lt;em&gt;withheld&lt;/em&gt;. That sounds backwards — why not just give the money back? Because an unrecorded credit is invisible to the ledger, and the ledger is the source of truth. A stuck debit that reconciliation will flag is recoverable; money that silently appeared in a wallet is not.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The gateway process dies between deduct and commit.&lt;/strong&gt; Every gateway periodically looks for pending intents that never committed. It claims each one with a compare-and-set lease, so only one gateway acts on it, reverses the debit, and emits a &lt;code&gt;VOIDED&lt;/code&gt; event with the original timestamp. The money is back in about a minute. Recovery is always &lt;em&gt;void&lt;/em&gt;, never &lt;em&gt;charge&lt;/em&gt; — the system never decides after the fact that you should have paid.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The creator fails.&lt;/strong&gt; A connection, DNS, or TLS failure or timeout voids the charge. A non-2xx response before the first byte streams is refunded in full. A 2xx without a valid proof header is voided too. But once the response has started streaming, the charge stands: the gateway is an immutable ledger, not a judge of whether a half-finished answer was useful.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;SQS delivers the same event twice.&lt;/strong&gt; It's a standard queue, so at-least-once delivery is a given. Each event has an ID of &lt;code&gt;correlation_id:status&lt;/code&gt;, the worker deduplicates per rollup bucket, and the same ID with a different body is treated as corruption and quarantined rather than guessed at.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The worker crashes mid-rollup.&lt;/strong&gt; A rollup posts as one database transaction under its own idempotency key, and SQS messages are deleted only &lt;em&gt;after&lt;/em&gt; that transaction commits. Redelivered events land as counted no-ops.&lt;/p&gt;

&lt;h2&gt;
  
  
  Idempotency without storing payloads
&lt;/h2&gt;

&lt;p&gt;Agents retry. Networks make them retry more. The &lt;code&gt;Idempotency-Key&lt;/code&gt; header is optional — requiring it would break the "no SDK" promise — but when it's present, the gateway binds it to a request hash: SHA-256 over a version tag, the key ID, the tool, the version, and the canonical request body with the JSON-RPC &lt;code&gt;id&lt;/code&gt; removed. So a retry that only changes the JSON-RPC id still matches.&lt;/p&gt;

&lt;p&gt;A retry with the same key gets &lt;code&gt;422 Idempotency_Conflict&lt;/code&gt;: no money moves and the tool doesn't run again. Notably, it does &lt;em&gt;not&lt;/em&gt; replay the original result, because the gateway never stores request or response payloads. That's a deliberate privacy trade — the gateway sees tool traffic but keeps none of it. The idempotency record itself lives for 120 hours.&lt;/p&gt;

&lt;p&gt;For any agent that retries, sending the header is the way to get exactly-once charging — and it costs one line of code, not an SDK.&lt;/p&gt;

&lt;h2&gt;
  
  
  A ledger that can't be quietly edited
&lt;/h2&gt;

&lt;p&gt;Redis is the authority at the moment of spending; PostgreSQL is the truth. Every five minutes, the worker rolls charges into a double-entry journal. Simplified:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight sql"&gt;&lt;code&gt;&lt;span class="k"&gt;CREATE&lt;/span&gt; &lt;span class="k"&gt;TABLE&lt;/span&gt; &lt;span class="n"&gt;transactions&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;
  &lt;span class="n"&gt;id&lt;/span&gt;              &lt;span class="kt"&gt;BIGSERIAL&lt;/span&gt; &lt;span class="k"&gt;PRIMARY&lt;/span&gt; &lt;span class="k"&gt;KEY&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;idempotency_key&lt;/span&gt; &lt;span class="kt"&gt;TEXT&lt;/span&gt; &lt;span class="k"&gt;NOT&lt;/span&gt; &lt;span class="kc"&gt;NULL&lt;/span&gt; &lt;span class="k"&gt;UNIQUE&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;request_hash&lt;/span&gt;    &lt;span class="kt"&gt;TEXT&lt;/span&gt; &lt;span class="k"&gt;NOT&lt;/span&gt; &lt;span class="kc"&gt;NULL&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;reversal_of&lt;/span&gt;     &lt;span class="kt"&gt;BIGINT&lt;/span&gt; &lt;span class="k"&gt;REFERENCES&lt;/span&gt; &lt;span class="n"&gt;transactions&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;posted_at&lt;/span&gt;       &lt;span class="kt"&gt;TIMESTAMPTZ&lt;/span&gt; &lt;span class="k"&gt;NOT&lt;/span&gt; &lt;span class="kc"&gt;NULL&lt;/span&gt; &lt;span class="k"&gt;DEFAULT&lt;/span&gt; &lt;span class="nb"&gt;now&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;

&lt;span class="k"&gt;CREATE&lt;/span&gt; &lt;span class="k"&gt;TABLE&lt;/span&gt; &lt;span class="n"&gt;entries&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;
  &lt;span class="n"&gt;transaction_id&lt;/span&gt; &lt;span class="kt"&gt;BIGINT&lt;/span&gt; &lt;span class="k"&gt;NOT&lt;/span&gt; &lt;span class="kc"&gt;NULL&lt;/span&gt; &lt;span class="k"&gt;REFERENCES&lt;/span&gt; &lt;span class="n"&gt;transactions&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;line&lt;/span&gt;           &lt;span class="kt"&gt;INT&lt;/span&gt;    &lt;span class="k"&gt;NOT&lt;/span&gt; &lt;span class="kc"&gt;NULL&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;account&lt;/span&gt;        &lt;span class="kt"&gt;TEXT&lt;/span&gt;   &lt;span class="k"&gt;NOT&lt;/span&gt; &lt;span class="kc"&gt;NULL&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="n"&gt;amount_micro&lt;/span&gt;   &lt;span class="kt"&gt;BIGINT&lt;/span&gt; &lt;span class="k"&gt;NOT&lt;/span&gt; &lt;span class="kc"&gt;NULL&lt;/span&gt; &lt;span class="k"&gt;CHECK&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;amount_micro&lt;/span&gt; &lt;span class="o"&gt;&amp;lt;&amp;gt;&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="k"&gt;PRIMARY&lt;/span&gt; &lt;span class="k"&gt;KEY&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;transaction_id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;line&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;

&lt;span class="c1"&gt;-- Every transaction must sum to zero. Checked once, at COMMIT,&lt;/span&gt;
&lt;span class="c1"&gt;-- so a transaction can insert its entries in any order.&lt;/span&gt;
&lt;span class="k"&gt;CREATE&lt;/span&gt; &lt;span class="k"&gt;CONSTRAINT&lt;/span&gt; &lt;span class="k"&gt;TRIGGER&lt;/span&gt; &lt;span class="n"&gt;entries_must_balance&lt;/span&gt;
  &lt;span class="no"&gt;AFTER&lt;/span&gt; &lt;span class="k"&gt;INSERT&lt;/span&gt; &lt;span class="k"&gt;ON&lt;/span&gt; &lt;span class="n"&gt;entries&lt;/span&gt;
  &lt;span class="no"&gt;DEFERRABLE&lt;/span&gt; &lt;span class="no"&gt;INITIALLY&lt;/span&gt; &lt;span class="no"&gt;DEFERRED&lt;/span&gt;
  &lt;span class="k"&gt;FOR&lt;/span&gt; &lt;span class="no"&gt;EACH&lt;/span&gt; &lt;span class="k"&gt;ROW&lt;/span&gt; &lt;span class="no"&gt;EXECUTE&lt;/span&gt; &lt;span class="k"&gt;FUNCTION&lt;/span&gt; &lt;span class="nf"&gt;assert_balanced&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;

&lt;span class="c1"&gt;-- Append-only: corrections are new reversing transactions, never edits.&lt;/span&gt;
&lt;span class="k"&gt;CREATE&lt;/span&gt; &lt;span class="k"&gt;TRIGGER&lt;/span&gt; &lt;span class="n"&gt;entries_append_only&lt;/span&gt;
  &lt;span class="no"&gt;BEFORE&lt;/span&gt; &lt;span class="k"&gt;UPDATE&lt;/span&gt; &lt;span class="k"&gt;OR&lt;/span&gt; &lt;span class="k"&gt;DELETE&lt;/span&gt; &lt;span class="k"&gt;ON&lt;/span&gt; &lt;span class="n"&gt;entries&lt;/span&gt;
  &lt;span class="k"&gt;FOR&lt;/span&gt; &lt;span class="no"&gt;EACH&lt;/span&gt; &lt;span class="k"&gt;ROW&lt;/span&gt; &lt;span class="no"&gt;EXECUTE&lt;/span&gt; &lt;span class="k"&gt;FUNCTION&lt;/span&gt; &lt;span class="nf"&gt;reject_mutation&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Truncation is blocked the same way, so the only way to change history is to add to it.&lt;/p&gt;

&lt;p&gt;Charges are bucketed by &lt;code&gt;unix(ts) / 300&lt;/code&gt;, and each bucket posts one balanced transaction per wallet, tool, and period. For a bucket where a wallet spent $1.20 on one tool with a 10% platform fee:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Account&lt;/th&gt;
&lt;th&gt;Side&lt;/th&gt;
&lt;th&gt;amount_micro&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Wallet liability&lt;/td&gt;
&lt;td&gt;Debit&lt;/td&gt;
&lt;td&gt;+1,200,000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Creator payable&lt;/td&gt;
&lt;td&gt;Credit&lt;/td&gt;
&lt;td&gt;−1,080,000&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Platform margin&lt;/td&gt;
&lt;td&gt;Credit&lt;/td&gt;
&lt;td&gt;−120,000&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Late events post a correction transaction rather than touching the old one. Posting uses &lt;code&gt;INSERT … ON CONFLICT DO NOTHING RETURNING&lt;/code&gt;, so the same key with the same hash replays harmlessly, and the same key with a &lt;em&gt;different&lt;/em&gt; hash is a loud error.&lt;/p&gt;

&lt;h3&gt;
  
  
  Reconciliation reports; humans repair
&lt;/h3&gt;

&lt;p&gt;A separate reconciliation tool compares the journal, the PostgreSQL wallet cache, and Redis, and classifies any drift. It is strictly read-only — it writes nothing to the journal, Redis, SQS, or Stripe — and exits non-zero for a person to investigate. Auto-repairing money is how one bug becomes two.&lt;/p&gt;

&lt;h2&gt;
  
  
  Signing what leaves the building
&lt;/h2&gt;

&lt;p&gt;The gateway calls servers it doesn't control, so creators need to know a request really came from Nexus402, and Nexus402 needs to not become a proxy into somebody's internal network.&lt;/p&gt;

&lt;p&gt;Every outbound request is signed with Ed25519 over six newline-joined fields: method, host, path and query, a Unix timestamp, the correlation ID, and a SHA-256 of the body. Creators accept a ±60-second window and verify against the host they &lt;em&gt;registered&lt;/em&gt;, not the one in the inbound header. The request is signed only after a connection slot is held, so a signature can't go stale waiting in a queue.&lt;/p&gt;

&lt;p&gt;On the way out, an SSRF guard blocks private, link-local, and cloud-metadata ranges, pins the resolved IP to defeat DNS rebinding, allows only HTTPS on port 443, never follows redirects, and refuses IPv4-mapped IPv6 literals — a classic way to sneak &lt;code&gt;127.0.0.1&lt;/code&gt; past a naive check.&lt;/p&gt;

&lt;h2&gt;
  
  
  The numbers, honestly
&lt;/h2&gt;

&lt;p&gt;I'll be precise here, because performance claims are where engineering writing most often lies.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Measured:&lt;/strong&gt; about &lt;strong&gt;210 requests per second&lt;/strong&gt; with a p50 of &lt;strong&gt;100 ms&lt;/strong&gt; end to end, from Kathmandu to AWS against real SQS. Most of that time is the publish and the creator round-trip, not gateway work.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Measured locally:&lt;/strong&gt; p50 17.4 ms and p99 33.3 ms against LocalStack, with Go's race detector on.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tested for correctness:&lt;/strong&gt; a double-spend gate runs 10,000 deductions against one wallet with 32 in flight, and the balance has to come out exact.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;&lt;a href="https://nexus402.dev" rel="noopener noreferrer"&gt;Nexus402&lt;/a&gt; is live — if you're building or calling paid MCP tools, take a look.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I'd tell someone building this
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Decide what failure costs before you write code.&lt;/strong&gt; Here, an outage may cost availability; it may never cost money. Every design choice above falls out of that one sentence.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Make the atomic unit genuinely atomic.&lt;/strong&gt; One script, one slot, one round-trip for the money decision. Anything spread across calls will eventually interleave.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Record before you reverse.&lt;/strong&gt; A reversal without a durable record is just a second bug.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Never float money, anywhere&lt;/strong&gt; — including inside Lua, inside JSON, and on your API.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Let reconciliation report, not repair.&lt;/strong&gt; The tool that finds drift should never be the tool that "fixes" it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Write down every decision.&lt;/strong&gt; Two hundred and fifty architecture decision records sounds excessive — until you need to remember, months later, exactly why the publish is synchronous.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If you're building payments for agents — or anything where "probably fine" isn't good enough — I'd like to compare notes. &lt;a href="https://nischaldhakal.com.np/contact" rel="noopener noreferrer"&gt;Get in touch&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>go</category>
      <category>agents</category>
      <category>redis</category>
      <category>postgres</category>
    </item>
  </channel>
</rss>
