Both trace agents, run offline and online evals, version prompts and offer a gateway for model calls. LangSmith goes deeper on the agent loop for LangChain and LangGraph teams; Logfire puts agent traces in the same OpenTelemetry view as the rest of a Python application, priced on records rather than seats.[1][4]
LangSmith is LangChain's platform focused on the agent development loop, while Pydantic Logfire is an OpenTelemetry observability platform that traces your whole application, HTTP, database and agent calls, with evals, prompt versioning and an AI gateway built in. LangSmith fits LangChain teams; Logfire fits Python teams, especially on Pydantic AI. Both are independent and both reserve self-hosting for Enterprise.
| Decision factor | LangSmith | Logfire |
|---|---|---|
| Ownership | LangChain, independent; $125M Series B in October 2025 | Pydantic, independent; $12.5M Series A led by Sequoia in 2024 |
| What it traces | LLM and agent runs, threads and subagent trajectories | The whole application: HTTP, database, LLM and agent spans |
| Online evals | Run by LangSmith through automation rules | Run in your application with pydantic-evals, then reported to Logfire |
| Human review | Annotation queues with rubrics and assigned reviewers | Run annotations in Beta; queues are early-access |
| Gateway | LLM Gateway (beta): spend caps, model access, redaction | AI gateway: guardrails that can flag, redact or block, spend caps, failover |
| Self-hosting | Enterprise add-on on Kubernetes; BYOC also Enterprise | Enterprise only, Kubernetes via Helm |
| Free tier | Developer: 1 seat, 5k base traces a month, 14-day retention | Personal: 1 seat, 10M records a month, 30-day retention |
| First paid tier | Plus: $39 per seat a month, 10k base traces | Team: $49 a month, 5 seats, 10M records, then $2 per million |
| Capability | LangSmith | Logfire | 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 | ✓ | Partial | ✓ |
| Prompt versioning and playground | ✓ | ✓ | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | Partial | ✓ |
| Gateway or proxy in the request path | — | ✓ | — |
| Licence and deployment | |||
| Open source | Partial | Partial | Partial |
| Self-hosting | ✓ | Partial | Partial |
This is the split that matters. Logfire is a general observability platform built on OpenTelemetry. It records the HTTP requests, database queries and background work around an agent in the same trace as the model and tool calls, with dedicated Agents and LLMs views on top. When an agent is slow because a downstream query is slow, you see both in one trace. Its richest auto-instrumentation and eval tooling are Python-first, and it comes from the team behind Pydantic and Pydantic AI.
LangSmith traces LLM and agent work: nested runs, threads for multi-turn sessions and a Trajectory view across subagents, opened to all users in September 2026. It does not try to replace your application monitoring. Its deepest integrations are with LangChain, LangGraph and Deep Agents, and it adds Engine, in public beta, which reads production traces and drafts fixes, and LangSmith Deployment for running agents.
Logfire's integration list is long: Pydantic AI, LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, Google ADK, Strands, Mastra, the Vercel AI SDK and more, and any standard OpenTelemetry SDK can send data. LangSmith covers many of the same agent frameworks, plus SDKs for Python, TypeScript, Go and Java.
Both support datasets, experiments, LLM judges and code scorers. The difference is where online evals execute. LangSmith runs online evaluators itself, through automation rules with filters and sampling, including on multi-turn threads, and puts flagged runs into annotation queues for reviewers.
Logfire's live evaluations run inside your application through the open-source pydantic-evals library, in the background at a sample rate you set, and emit OpenTelemetry events that Logfire aggregates. That keeps judge calls and their cost in your own code and accounts, but it means your service does the work. Logfire's run annotations are in Beta and annotation queues are an early-access feature, so structured human review is less developed.
Prompt versioning is close on both. Logfire saves immutable prompt versions, tests them against saved scenarios and datasets, and promotes a version through labels with rollout targeting. LangSmith's Prompt and Context Hub uses commit-based versions and tags, with a playground.
Both vendors now sit optionally in the model-call path. Logfire's AI gateway, the Pydantic AI Gateway folded into Logfire in March 2026, routes to several providers with failover and load balancing, enforces spend caps, and runs guardrails that can observe, flag, redact or block prompts and completions containing secrets or personal data. LangSmith's LLM Gateway, in public beta since 30 July 2026, enforces spend caps, model access policies and redaction, with custom regex rules from October 2026.
Neither gateway covers an agent's tool calls, and neither documents approvals on agent actions. LangChain describes tool and MCP gateways as planned.
Logfire's gateway is offered on its cloud plans and on self-hosted Logfire, while LangSmith's is still in beta.
As of 20 October 2026, LangSmith charges per seat plus traces: Developer is free for one seat and 5k base traces a month, and Plus is $39 per seat a month with 10k base traces, with the per-trace overage rate unpublished. Logfire charges on records, meaning spans, logs and metrics: Personal is free for one seat and 10M records a month, Team is $49 a month with five seats and 10M records, Growth is $249 a month with unlimited seats, and extra records cost $2 per million.
Because Logfire counts every span in the application, not only agent spans, estimate your full record volume. Check LangSmith's pricing and Logfire's pricing against your own numbers.
Retention differs: 30 days on Logfire's Personal and Team plans and up to 90 days on Growth, against 14 days for LangSmith base traces, with extended retention up to 180 days. Logfire also lets you set a spending cap on paid plans.
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 and acts on what it finds through runtime policies on tool calls and data access, approval holds for risky actions, and a kill switch for a run, an agent or a team. Both LangSmith and Logfire act only on model calls routed through their gateways; neither holds a tool action for a person. Prefactor does not trace your wider application as Logfire does, has no prompt playground or prompt registry, and is hosted. 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.