Guides / culpa vs agentwatch
Culpa vs AgentWatch
AgentWatch is an edge gateway that detects a spiralling agent, blocks it with a 402, and keeps replay and cross-provider forensics on the traffic it routes. Culpa, a local-first LLM cost, margin, and forecast ledger, stops nothing. It runs on your own infrastructure, adds margin against revenue, and forecasts the next month.
Why this happens
Both watch runaway agents and both report on spend, so the real question is where each one sits. AgentWatch is a gateway. Your traffic routes through it, which is what lets it refuse a call mid-loop, and it keeps the forensics on that traffic at its own edge. Culpa never sits in the request path and can't block anything, and in exchange the data stays on your infrastructure and it captures calls that never go near a gateway. The second difference is scope. AgentWatch answers what your agents spent. Culpa carries revenue too, so it answers whether that customer is profitable, and forecasts the next month.
What this usually looks like
- An agent looped overnight and the first anyone knew was the invoice.
- You have a spend ceiling and still can't say which feature consumed the budget.
- Nobody can tell you what next month costs at current growth.
- A customer's usage doubled and nobody knows whether they're still profitable.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Mistakes that cost the most
| Mistake | Why it hurts | Do instead |
|---|---|---|
| Expecting spend reporting to answer a margin question. | Knowing a customer cost $80 decides nothing until you set it against what they pay you. | Check which tool holds revenue. A cost figure alone can't tell you whether an account is worth keeping. |
| Expecting a ledger to stop a runaway agent. | Culpa reports rather than blocks, so a loop that starts at 2am runs until something stops it. | If you need a hard stop, run something that enforces one. Culpa is the record, not the brake. |
| Assuming any cost layer keeps your prompts. | It rules out useful tools on a concern that may not apply, and both of these publish their answer. | Read each vendor's retention statement. AgentWatch publishes zero prompt retention and Culpa keeps prompts on your own infrastructure. |
Run this check tonight
- Write down the last incident. Did you need it stopped, or explained?
- If stopped: does anything in your stack enforce a hard ceiling today?
- If explained: can anything name the conversation and the customer behind the spend?
- Check whether either layer sits in the request path, and what happens when it fails.
- Most teams find they answered yes to both questions and have only one of the two.
The same runaway loop, seen by a ceiling and by a ledger
AgentWatch's published behaviour and plans, read from agent-watch.dev on 2026-08-03, against what Culpa does with the same incident. No pricing comparison is drawn: AgentWatch prices per request and Culpa prices per tracked call, so a like-for-like figure would be invented.
Both explain spend and only one can refuse a call. The choice is really about where the data sits and how far the question goes: a gateway that can stop a loop, or a ledger on your own infrastructure that carries revenue and a forecast. A team burned by a loop often wants both.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $39 to $99 per month | AgentWatch's published Starter and Pro plan prices | provider-reported | agent-watch.dev pricing, read 2026-08-03. Starter covers 150,000 requests a month and Pro 500,000, with a free tier at 50,000. |
What a generic answer can’t know
Whether you need a brake or a record depends on which incident you actually had, and only your own history says. Run the free scan on your existing spend and see whether the conversation it names is one you'd have found. Culpa keeps your prompts and responses on your infrastructure, and counts the calls to run your plan.
Questions founders ask next
Does Culpa block runaway agents like AgentWatch does?
No, and that's a real limitation rather than a positioning choice. Culpa is a ledger and a forecast, not an enforcement layer, because it never sits in the request path. A loop that starts at 2am runs until something else stops it. If you want a hard stop, AgentWatch's ceiling and kill-switch are built for exactly that.
Does either one keep my prompts?
AgentWatch's site states it never stores prompts or completions and keeps only metadata such as tokens, latency and cost. Culpa runs on your own infrastructure, so your prompts and responses stay there and Culpa counts the calls to run your plan. Both published answers are read 2026-08-03.
Will either break my app if it goes down?
Both are designed not to. AgentWatch states it fails open by default, so calls proceed if it has a problem. Culpa's capture paths fail open by contract, and the in-process route never sits between your app and the provider at all.
Can they run together?
Yes, and it's a sensible pairing. A ceiling keeps a loop from becoming an invoice, and a ledger tells you which feature and customer the surviving spend belongs to.
On your infrastructure
Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.
How Culpa works
Find the culprit. Not just the total.
Your dashboard shows what you spent. It stops short of who spent it. Culpa shows the conversation, the user and the feature behind it.
Your prompts stay local.
Culpa runs on your own infrastructure. What you send to a model reaches us at no point.
Every dollar has a name.
Follow any charge to the conversation, the user, the feature and the customer behind it.
See the bill before it lands.
Cost your next feature before you ship it. You get the likely bill and the worst case, at best, median, p90 and p99.
Three steps to your first answer.
Change one base URL.
Or drop in the Python or TypeScript library.
Find your most expensive conversation.
In the first session, not the first week.
Cost your next feature before you ship it.
Why the bill went up
Example dashboardCalls traced
418,209
across 3 projects
Spend this week
$378.41
+ $182 vs last week
Failed calls
312
74% retried, and you paid for all of them
+ $182 this week traced to one culprit
Spend over 14 days
Most expensive users
Next week forecast
Graded against reality. Accuracy shown as results land.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Keep reading
Sources: AgentWatch. Last reviewed 2026-08-03. Plain text version.