Guides / culpa vs langsmith
Culpa vs LangSmith
LangSmith is LangChain's tracing and evaluation platform, and it self-hosts inside your own VPC on the Enterprise plan, so traces stay on your infrastructure. Culpa, a local-first LLM cost, margin, and forecast ledger, does that by default rather than by contract. The comparison that decides is elsewhere though: Culpa carries the revenue a customer pays you and forecasts the next month, and LangSmith claims neither.
Why this happens
Privacy is the argument everyone expects here and it barely separates these two. LangSmith publishes managed cloud, bring-your-own-cloud and self-hosted options, and says plainly that self-hosting keeps sensitive traces off its infrastructure. The one real difference is how you get it: LangSmith's self-host is an Enterprise add-on behind a sales conversation and a licence key, where Culpa runs locally by default. That's a procurement difference rather than a capability one. What actually differs is scope. LangSmith answers what a run did and whether it was correct, and it groups cost by user or thread while it's at it. It holds no revenue, so it can total a customer's spend and still can't say whether that customer is worth keeping, and it publishes no forecast, so last month is the only month.
What this usually looks like
- You can total a customer's spend and can't say whether they're profitable.
- Nobody has put a customer's cost beside what that customer pays you.
- A forecast for next month exists only as last month plus a feeling.
- Traces are rich and the finance question still goes to a spreadsheet.
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 |
|---|---|---|
| Choosing a cost tool on the privacy argument alone. | Most of this category self-hosts, so privacy rarely separates the shortlist. | Ask which tool holds revenue and which publishes a forecast. That thins the list fast. |
| Reading spend per customer as margin per customer. | Spend is one side of it, and the side that decides is what they pay you. | Put the two together. A customer at $80 of cost is fine or fatal depending on their plan. |
| Assuming a tracing tool will grow into a finance tool. | Traces and revenue are different data with different owners, and few vendors cross that line. | Check what each tool claims today rather than what its roadmap implies. |
Run this check tonight
- Ask your current tool what one named customer cost you last month.
- Then ask it what that customer paid you. Most stop at the first question.
- Ask it for next month's spend as a range rather than a point.
- Check whether your traces and your revenue live in the same place at all.
- Whatever needs a spreadsheet to answer is the gap.
What each one publishes, read off both sites
LangSmith's own product and deployment pages, read 2026-08-03, against what Culpa does. No pricing comparison is drawn: LangSmith prices per trace and seat and Culpa per tracked call, so a like-for-like figure would need an assumed traffic mix to mean anything.
On privacy these two agree, and a comparison page that pretended otherwise would be selling rather than helping. The real question is whether you need the finance half, and only one of them offers it.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $80.00 per customer per month | modelled monthly model spend for one customer, the number a tracing tool can produce and a margin question still needs more than | calculated | 40M input and 8M output tokens a month on Claude Haiku 4.5 at real rates of $1.00 and $5.00 per million from the price book, effective 2026-07-02: (40 x $1.00) + (8 x $5.00) = $40.00 + $40.00 = $80.00. Token volumes are modelled. The point of the figure is that it settles nothing until it sits beside what that customer pays. |
What a generic answer can’t know
Whether a customer is worth keeping needs their cost and their price in the same view, and no tracing tool holds the second one. That join is yours to make. Culpa makes it on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.
Questions founders ask next
Is Culpa more private than LangSmith?
Not in capability. LangSmith publishes a self-hosted option and says traces then never leave your infrastructure, which is the property Culpa leads with. The difference is how you get there: LangSmith gates self-hosting behind its Enterprise plan and a licence key, where Culpa runs locally by default. Privacy is close to table stakes in this category rather than a differentiator.
What does Culpa do that LangSmith doesn't claim?
Two things, checked against LangSmith's own pages on 2026-08-03. Margin per customer, which needs the revenue that customer pays you, and a forecast of next month from your own history. Neither appears in its published feature set. Note what's absent from that list: LangSmith does group cost by user and by thread, so cost attribution isn't the difference.
Can I run both?
Yes, and it's a reasonable stack. Tracing answers why a run behaved as it did and a ledger answers what your runs cost and whether the customer behind them is profitable. They read the same traffic for different reasons.
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: LangSmith, Self-hosted LangSmith, LangChain pricing. Last reviewed 2026-08-03. Plain text version.