Both trace agents, run experiments, score production traffic, queue runs for human review and version prompts. LangSmith is the first-party choice for teams on LangChain or LangGraph; Braintrust is framework-agnostic and organised around experiments, with unlimited users on every plan. Your framework and your team size usually decide it.[1][4]
Braintrust is a framework-agnostic hosted platform built around experiments, while LangSmith is LangChain's platform built around traces, with its deepest integration in LangChain and LangGraph. Both are proprietary, both independent, and both reserve self-hosting for Enterprise. Braintrust includes unlimited users on every plan; LangSmith charges per seat.
| Decision factor | Braintrust | LangSmith |
|---|---|---|
| Ownership | Independent; $80M Series B in February 2026 | LangChain, independent; $125M Series B in October 2025 |
| Centre of gravity | Experiments: dataset, task, scorers, compare | Traces: instrument, curate datasets from production, evaluate |
| Framework coupling | Framework-agnostic, with integrations for most agent frameworks | Works standalone; deepest with LangChain, LangGraph and Deep Agents |
| Analysis agent | Loop, plus Patterns and Topics (several in public preview) | Engine (public beta), which drafts proposed fixes from traces |
| Gateway | Braintrust Gateway: public preview, free, no SLA | LLM Gateway (beta): spend caps, model access and redaction on model calls |
| Self-hosting | Enterprise only, data plane only; Braintrust runs the control plane | Enterprise add-on, full deployment on Kubernetes; BYOC also Enterprise |
| Free tier | Starter: 1 GB processed data, 10K scores, 14-day retention, unlimited users | Developer: 1 seat, 5k base traces a month, 14-day retention |
| First paid tier | Pro: $249 a month, unlimited users | Plus: $39 per seat a month, 10k base traces |
| Capability | Braintrust | LangSmith | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | ✓ | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | ✓ | ✓ |
| Code-based evaluators | ✓ | ✓ | ✓ |
| Human review and annotation | ✓ | ✓ | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | — | Partial | ✓ |
| Gateway or proxy in the request path | ✓ | — | — |
| Licence and deployment | |||
| Open source | Partial | Partial | Partial |
| Self-hosting | Partial | ✓ | Partial |
Feature lists for the two now overlap almost completely. What still separates them is where the work starts. Braintrust starts from the experiment: you bring a dataset, define a task and scorers, run variants and compare. Its Loop agent reads results, writes scorers and suggests changes, and since September 2026 Patterns group production behaviour for review.
LangSmith starts from the trace. You instrument, production traffic flows in, you add interesting runs to datasets, and evaluations grow from what you see. Its Engine, in public beta since May 2026, reads production traces, opens issues and drafts fixes.
Teams that are pre-launch and iterating on prompts and models often find the experiment loop faster. Teams already in production whose first problem is understanding what agents do often start from traces. Both products now cover the other motion, so this is a tiebreaker rather than a hard limit.
A practical test during a trial: bring one agent run that failed part way through and see which tool points you to the step that broke. Both now offer trajectory views; Braintrust also published behaviour specs, an open standard for judging agent trajectories, in July 2026.
LangSmith's integration advantage is real. The same company builds the frameworks, so instrumentation follows their abstractions, and the Trajectory view, opened to all users in September 2026, combines messages across a main agent and its subagents. LangSmith Deployment, formerly LangGraph Platform, runs those agents from the same vendor.
Braintrust also integrates with LangChain and LangGraph, alongside the OpenAI Agents SDK, Claude Agent SDK, CrewAI, Google ADK, Mastra, Pydantic AI and others, and both accept OpenTelemetry traces. Off the LangChain stack, LangSmith's advantage narrows and the decision moves to workflow, pricing units and hosting. Check how portable traces and datasets are before you are deep in either; LangSmith's SDK can export to other OpenTelemetry backends, which helps.
Both now offer a gateway in the model-call path, and both are pre-release. Braintrust Gateway gives one API to OpenAI, Anthropic, Google, AWS and others with caching and logging; the hosted version is in public preview with no SLA. LangSmith's LLM Gateway, in public beta since 30 July 2026, adds spend caps, model access policies and redaction of personal data and secrets. Neither gateway covers tool calls, and Braintrust documents no blocking or approvals on live agent actions.
Self-hosting is Enterprise only on both, but the scope differs. Braintrust lets you run the data plane in your own AWS, GCP or Azure account while it always operates the control plane, including the UI and authentication. Self-hosted LangSmith is a full deployment on Kubernetes. If a compliance review requires everything inside your network, that difference matters.
As of 20 October 2026, Braintrust charges a platform fee plus processed data and scores, with unlimited users on every plan: Starter is free with 1 GB and 10K scores, and Pro is $249 a month with 5 GB and 50K scores. LangSmith charges per seat plus traces: Developer is free for one seat and 5k base traces, and Plus is $39 per seat a month with 10k base traces. LangChain does not publish its per-trace overage rate.
For a small team LangSmith Plus costs less. At seven seats or more, Plus seat fees alone pass Braintrust Pro's monthly fee, before usage on either side. Check Braintrust's pricing and LangSmith's pricing against your own seat count and volumes.
Retention differs too. Braintrust Starter keeps data for 14 days and Pro for 30, extendable to 180 days at $0.50 per GB a month. LangSmith keeps base traces for 14 days, with extended retention capped at 180 days on its cloud service.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of a run, including tool calls, and acts on what it finds through runtime policies, approval holds for risky actions and a kill switch for a run, an agent or a team. Braintrust scores after logging and has no run-time controls; LangSmith's gateway can cap spend and redact model calls but does not touch tool calls. Prefactor has no prompt playground or prompt registry and its platform is hosted, so for a prompt iteration loop Braintrust or LangSmith is the better pick. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.