Best LLM observability tools in 2026
Trace prompts, model calls and agent steps, and track cost, latency and output quality of AI applications. This list has the top 14 of 22 LLM observability tools, ranked by a published score. Of these, 7 are open source, 14 have a free plan and 14 can be self-hosted. Checked by hand on 30 September 2026.
How this list is ranked
Each active tool gets a score out of 100. Archived, end-of-sale and maintenance-only tools are left out.
- Capabilities: up to 75 points for the share of the category checklist that the tool's own documentation confirms. A tool meeting 8 of 10 items gets 60.
- Free to start: 15 points for a free plan or a free edition.
- Open or self-hostable: 10 points if it is open source or can be self-hosted.
Each category has a fixed checklist of capabilities (shown in the table below and on every tool page). An item counts only when the tool's own documentation confirms it; anything we could not confirm counts as not met. How long a tool has existed is deliberately not scored, so a new tool is not ranked below an older one for being new. Tools with the same score share a rank (shown as =), are listed A to Z within it, and the list never cuts between them. Nobody pays for a position.
- =1
Braintrust
Evaluation and observability platform for AI products: logs production traces, runs evals and scorers, and offers a prompt playground and datasets.
100 10 of 10 capabilities (75), free 15, open 10 - =1
Future AGI
Open source platform to trace, evaluate, simulate and guard LLM and agent apps, including an AI gateway.
100 10 of 10 capabilities (75), free 15, open 10 - =1
Galileo
AI evaluation and observability platform with tracing, evaluation metrics and real-time guardrails for agents; hosted, VPC or on-prem.
100 10 of 10 capabilities (75), free 15, open 10 - =1
HoneyHive
Observability and evaluation platform for AI agents: OpenTelemetry-based tracing, automated and human evals, prompt versioning and production monitoring.
100 10 of 10 capabilities (75), free 15, open 10 - =1
LangSmith
LangChain's platform for tracing, evaluating and monitoring LLM apps and agents, usable with or without LangChain; also hosts agent deployments.
100 10 of 10 capabilities (75), free 15, open 10 - =1
LangWatch
Open source LLMOps platform: traces LLM and agent runs, runs evaluations and agent simulations, and helps optimize prompts; cloud or self-managed.
100 10 of 10 capabilities (75), free 15, open 10 - =1
Opik
Open source LLM and agent observability from Comet: tracing, test suites, evaluations and a prompt optimizer; self-host the same code as the cloud.
100 10 of 10 capabilities (75), free 15, open 10 - =1
PostHog
Open source product suite with session replay, product analytics, feature flags, error tracking and LLM analytics sharing one event store. Hosted with a monthly free allowance, or self-hosted.
100 10 of 10 capabilities (75), free 15, open 10 - =9
Arize Phoenix
Source-available LLM tracing and evaluation tool built on OpenTelemetry and OpenInference; run it locally, in a notebook or on your servers to inspect traces and score outputs.
93 9 of 10 capabilities (68), free 15, open 10 - =9
Confident AI
Evaluation and observability platform from the makers of the open source DeepEval library: LLM tracing, regression tests in CI, and red teaming.
93 9 of 10 capabilities (68), free 15, open 10 - =9
Laminar
Open source observability for AI agents: OpenTelemetry-based tracing, evaluations and analysis of agent runs, self-hostable or cloud.
93 9 of 10 capabilities (68), free 15, open 10 - =9
Langfuse
Open source LLM engineering platform: traces LLM and agent calls, manages prompts, runs evaluations and tracks cost and latency. Self-host or use the cloud; acquired by ClickHouse in 2026.
93 9 of 10 capabilities (68), free 15, open 10 - =9
OpenLLMetry
Open source OpenTelemetry instrumentations for LLM providers, vector databases and agent frameworks, by Traceloop; sends traces to any OTLP backend or the Traceloop platform.
93 9 of 10 capabilities (68), free 15, open 10 - =9
W&B Weave
Weights & Biases toolkit for tracing LLM calls, running evaluations with custom scorers and comparing model and prompt versions.
93 9 of 10 capabilities (68), free 15, open 10
Side by side
| Tool | Pricing | Free plan | Open source | Self-hosted | LLM call tracing | Agent and multi-step traces | Token and cost tracking | Prompt management | Evaluations | Datasets and experiments | User feedback | OpenTelemetry support | Playground | Alerts |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Braintrust | Free plan | 1 GB processed data, 10k scores, 14-day retention | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Future AGI | Free plan | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Galileo | Free plan | Yes | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| HoneyHive | Free plan | 10,000 events/mo, 5 users, 30-day retention | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| LangSmith | Free plan | 5k base traces/mo, 1 seat | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| LangWatch | Free plan | 50k events/mo, 14-day data access, 2 users | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Opik | Free plan | 25k spans/mo, 10 members, 60-day retention | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| PostHog | Free plan | 5K session recordings and 1M events per month | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Arize Phoenix | Free | Yes | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | - |
| Confident AI | Free plan | 2 seats, 1 project, 1 GB-month of trace spans | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | - | Yes |
| Laminar | Free plan | 1 GB data, 7-day retention, 1 seat | Yes | Yes | Yes | Yes | Yes | - | Yes | Yes | Yes | Yes | Yes | Yes |
| Langfuse | Free plan | 50k units/mo, 30 days data access, 2 users | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | - |
| OpenLLMetry | Free plan | Traceloop: 50K spans/mo, 5 seats, 24h retention | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | - |
| W&B Weave | Free plan | Yes | No | Yes | Yes | Yes | - | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
Yes: confirmed in the tool's documentation. No: documented as not offered. -: not confirmed. Plans and prices change; check each tool's own site before you decide.
More lists
- LLM observability tools in 2026 22
- Open source LLM observability tools in 2026 11
- Self-hosted LLM observability tools in 2026 20
- Best observability tools in 2026 47
- Best observability platforms in 2026 11
- Best metrics and time-series tools in 2026 5
- Best log management tools in 2026 11
- Best tracing and APM tools in 2026 10
- Best error tracking tools in 2026 11
- Best continuous profiling tools in 2026 5
- Best real user monitoring and session replay tools in 2026 12
- Best OpenTelemetry and telemetry pipeline tools in 2026 12
- Best dashboard tools in 2026 5