Skip to content
SPECIFICTION
Get listed

OpenLIT

openlit.io

Free plan Open source WebWindowsmacOSLinuxAPISelf-hosted
Visit openlit.io
Spec score 8.5 #9 of 31 in AI Agent Observability Tools

What it is

OpenLIT is an open-source platform for tracing and evaluating LLM activity, managing prompts, and tracking costs. Built on OpenTelemetry, it can capture model calls, tool use, and retrievals, with filters for user, session, cost, latency, and metadata. Evaluations can use an LLM judge, heuristics, or human review on production traffic or experiments. Prompt Hub provides versioning, dynamic variables, API retrieval, unlimited prompt fetching, and AI-assisted improvement. OpenGround supports side-by-side model comparisons for latency, tokens, and cost, including comparisons using a user’s own key. Coding agent monitoring covers sessions, costs, tokens, repository activity, and code impact. The site lists more than 56 integrations across LLMs, agent frameworks, vector databases, and GPUs. OpenLIT accepts OTLP data and can be self-hosted with Docker Compose or Kubernetes Helm. Its Apache 2.0 license and site description note that self-hosting keeps prompts, traces, and secrets in the user’s environment. The OSS plan costs nothing and includes unlimited self-hosted usage, users, projects, and environments.

Spec sheet

What it does
OpenLIT is an OpenTelemetry-native platform for tracing, evaluating, guarding and improving AI agents and coding agents.Read 14 Oct 2026 · openlit.io
Deployment
The platform can be self-hosted using Docker Compose or Kubernetes Helm and accepts OpenTelemetry data through OTLP.Read 29 Sep 2026 · openlit.io
Integrations
The home page advertises more than 56 integrations spanning LLMs, agent frameworks, vector databases and GPUs.Read 14 Oct 2026 · openlit.io
Support
The OSS plan includes community support on GitHub.Read 29 Sep 2026 · openlit.io
Agent observability
The platform records LLM calls, tool calls and agent steps for production analysis.Read 14 Oct 2026 · openlit.io
Coding-agent monitoring
OpenLIT tracks coding-agent sessions, costs, tokens and repository activity.Read 14 Oct 2026 · openlit.io
Company background
OpenLIT says the project is community-driven, built in public on GitHub, and used by developers worldwide.Read 29 Sep 2026 · openlit.io
Company founder
Crunchbase identifies Aman Agarwal as a founder.Read 14 Oct 2026 · crunchbase.com
Company headquarters
Crunchbase places OpenLIT in New Delhi, Delhi, India.Read 14 Oct 2026 · crunchbase.com
Company legal name
Crunchbase lists the legal name as OpenLIT and operating status as active.Read 14 Oct 2026 · crunchbase.com
Company size
Crunchbase lists the company size as 1–10 employees.Read 14 Oct 2026 · crunchbase.com
Company type
Crunchbase classifies OpenLIT as a private non-profit company.Read 14 Oct 2026 · crunchbase.com
Custom dashboards
Dashboards are backed by ClickHouse and support charts, tables, filters, folders, import and export.Read 14 Oct 2026 · openlit.io
Documentation depth
The documentation includes provider-specific SDK guides, observability pages, controller deployment material, CLI configuration and an indexed documentation file.Read 14 Oct 2026 · docs.openlit.io
Evaluations
OpenLIT includes built-in evaluations, programmatic evaluations and LLM-as-a-judge workflows that can run locally or in CI.Read 14 Oct 2026 · openlit.io
Funding disclosure
The reviewed Crunchbase profile does not show a funding total or disclosed funding rounds for OpenLIT.Read 14 Oct 2026 · crunchbase.com
Go SDK capability
The Go SDK wraps provider clients and emits traces and metrics for standard and streaming calls, embeddings and image generation.Read 14 Oct 2026 · docs.openlit.io
GPU monitoring
The GPU collector and dashboards expose utilization, memory, temperature and power metrics.Read 14 Oct 2026 · openlit.io
Guardrails
The SDK includes guards for prompt injection, jailbreak and sensitive-topic detection using LLM and regular-expression checks.Read 14 Oct 2026 · openlit.io
Integration examples
Documented integrations include OpenAI, Anthropic, Cohere, Mistral, Ollama, Bedrock, Vertex AI, LangChain, LlamaIndex, Pinecone, Chroma, Qdrant and Milvus.Read 14 Oct 2026 · openlit.io
Licence
The project is licensed under Apache License 2.0.Read 14 Oct 2026 · github.com
License and data
OpenLIT is Apache 2.0 licensed, and its site says self-hosting keeps prompts, traces, and secrets in the user's environment.Read 29 Sep 2026 · openlit.io
Limit: independent review coverage
Product Hunt lists OpenLIT launches but currently shows no user reviews.Read 14 Oct 2026 · producthunt.com
Model comparison
OpenGround compares models side by side for latency, tokens, and cost, and supports bring-your-own-key model comparison.Read 29 Sep 2026 · openlit.io
Notable limits
Fleet Hub lists single-cluster collector monitoring, while the self-hosted plan lists unlimited usage, historical data, users, projects, and environments.Read 29 Sep 2026 · openlit.io
Notable news: integration expansion
An OpenLIT LinkedIn post announced integrations with more than 50 AI tools during its launch week, including new model, agent and vector-database connectors.Read 14 Oct 2026 · linkedin.com
Notable news: Product Hunt launch
Product Hunt records OpenLIT 2.0 launching January 16, 2025 and the zero-code observability launch on October 10, 2025.Read 14 Oct 2026 · producthunt.com
OpenGround
OpenGround compares models side by side using latency, token and cost measurements with bring-your-own keys.Read 14 Oct 2026 · openlit.io
OpenTelemetry routing
Telemetry uses standard OTLP and can be exported to Grafana, Datadog, New Relic, SigNoz, Jaeger or another compatible backend.Read 14 Oct 2026 · openlit.io
Package release
PyPI lists openlit version 1.45.0 uploaded on August 3, 2026, with Python wheels and source distributions.Read 14 Oct 2026 · pypi.org
Prompt Hub
Prompt Hub supports prompt versioning, dynamic variables, API retrieval, unlimited prompt fetching, and AI-assisted improvement.Read 29 Sep 2026 · openlit.io
Prompt management
Prompt Hub provides prompt versioning, dynamic variables, API retrieval and AI-assisted prompt improvement.Read 14 Oct 2026 · openlit.io
Public API
The sandbox exposes an OpenAPI description and API-key scopes covering ingestion, telemetry, prompts, vault, evaluations, chat, controller and database configuration.Read 14 Oct 2026 · openlit.io
Rating on G2
5.0 out of 5 from 1 review.Read 14 Oct 2026 · g2.com
SDKs
Official SDKs are available for Python, TypeScript and Go; Python installs with pip and TypeScript with npm.Read 14 Oct 2026 · github.com
Secret storage
Vault provides encrypted storage for LLM API keys using AES-256-GCM, with key and tag based retrieval and API authentication.Read 29 Sep 2026 · openlit.io
Security detail
CLI configuration files are written with 0600 permissions and API keys are not passed into launched agent child processes.Read 14 Oct 2026 · github.com
Security sign-in
The OSS plan lists sign-in with Google, Azure AD, and GitHub.Read 29 Sep 2026 · openlit.io
Self-hosting
Docker Compose runs the UI on port 3000, OTLP receivers on ports 4317 and 4318, and ClickHouse for traces, metrics and logs; Kubernetes Helm deployment is also documented.Read 14 Oct 2026 · openlit.io
System requirements
The local sandbox requires Docker and Docker Compose; Kubernetes users install the published Helm chart.Read 14 Oct 2026 · openlit.io
Tracing
LLM tracing can capture calls, tools, and retrievals, with filters for user, session, cost, latency, and metadata.Read 29 Sep 2026 · openlit.io
Vault
Vault stores LLM API keys in encrypted AES-256-GCM storage and supports key/tag retrieval and API authentication.Read 14 Oct 2026 · openlit.io
Zero-code controller
The Controller uses eBPF and SDK injection to instrument Kubernetes, Docker and Linux workloads without code changes, image rebuilds or redeployment.Read 14 Oct 2026 · docs.openlit.io

Plans and prices

OSS Free Unlimited self-hosted usage · unlimited users · unlimited projects and environments · community support (GitHub) Read 29 Sep 2026 · openlit.io
OSS Free Unlimited self-hosted usage · unlimited users · unlimited projects/environments · community GitHub support · Apache 2.0 Read 14 Oct 2026 · openlit.io
Cloud Priced on request Fully hosted service · managed infrastructure and upgrades · pricing shared at launch · waitlist only Read 14 Oct 2026 · openlit.io

Every OpenLIT plan and price →

Strong points and limits

Strong points

  • Captures calls, tools, and retrievals with filters.
  • Evaluations include LLM, heuristic, and human review.
  • Prompt Hub includes versioning and AI-assisted improvement.
  • OSS plan includes unlimited self-hosted usage.
  • Supports Docker Compose, Kubernetes Helm, and OTLP.

Limits

  • Cloud plan is coming soon; price is unlisted.
  • Fleet Hub lists single-cluster collector monitoring.
  • OSS support is community support on GitHub.

Who it suits

OpenLIT suits teams that need LLM tracing, evaluation, prompt management, and cost tracking, especially those seeking self-hosted deployment. Its listed capabilities also address coding-agent monitoring and teams using multiple LLM and infrastructure integrations.

OpenLIT in depth

Read the full write-up

OpenLIT is an open-source platform for observing and improving LLM and agent systems. It is a strong fit for engineering teams comfortable running infrastructure themselves, especially those already using OpenTelemetry. Its broad tracing, evaluation, prompt and cost tooling is compelling, but teams wanting a managed service must look elsewhere for now.

Overview

OpenLIT brings operational visibility and development controls into one Agent Harness Engineering platform. Built on OpenTelemetry, it records model calls, tool use and retrievals, then connects that data to evaluations, prompt workflows, model comparisons and spending analysis. That breadth suits teams building both conventional LLM applications and multi-step agents, rather than buyers seeking a narrowly focused request logger.

The Apache 2.0 license permits a self-managed deployment with Docker Compose or Kubernetes Helm. OpenLIT says this arrangement keeps prompts, traces and secrets inside the user's environment, a meaningful advantage for organizations with strict data boundaries. The trade-off is operational ownership: there is no currently hosted Cloud service to take deployment and maintenance off the team's hands.

Key features

Tracing and evaluation

Tracing captures calls, tools and retrievals, with filters for user, session, cost, latency and metadata. Those dimensions make it practical to isolate a slow session, an expensive user journey or a particular retrieval path. Evaluations support LLM-as-a-judge, heuristic checks and human review on production traffic or experiments, so teams can combine automated regression checks with manual assessment instead of relying on a single scoring method.

Prompt and model workflows

Prompt Hub provides versioning, dynamic variables, API retrieval, unlimited prompt fetching and AI-assisted improvement. It is suited to teams that need prompts managed as reusable application assets rather than embedded in code. OpenGround compares models side by side on latency, token use and cost, and supports comparisons with a customer's own model keys; that makes trade-offs visible before standardizing on a provider.

Agent and infrastructure coverage

Coding-agent observability follows sessions, cost, tokens, repository activity and code impact, giving engineering-focused teams context beyond model metrics. More than 56 integrations span LLM providers, agent frameworks, vector databases and GPUs, including OpenAI, Anthropic, LangChain, Pinecone and NVIDIA NIM. Vault stores LLM API keys with AES-256-GCM encryption and retrieves them through keys, tags and authenticated APIs. OTLP ingestion and the Docker Compose and Helm options make it adaptable to existing telemetry pipelines.

Pricing

OSS

The OSS plan costs nothing and includes unlimited self-hosted usage, users, projects and environments. It also includes community support through GitHub, with sign-in via Google, Azure AD or GitHub. This is the practical choice for teams able to run their own deployment and comfortable using community support instead of a contracted service desk. Fleet Hub's single-cluster collector monitoring is a stated limit, although the self-hosted plan otherwise removes usage, user, project and environment caps.

Cloud

Cloud is coming soon and uses custom pricing. It is intended for buyers who want fully hosted operation, but its feature set and commercial terms are not yet defined, so it cannot currently be evaluated as an available purchasing option.

Platforms

OpenLIT supports API, Linux, macOS, Windows, web and self-hosted deployments. The self-hosted route is the central deployment model, using Docker Compose or Kubernetes Helm and accepting OpenTelemetry data over OTLP. That range works for teams instrumenting services through an API as well as operators managing the platform directly.

Who it's for

Choose OpenLIT if your team needs tracing, evaluations, prompt management and token-cost tracking in one open-source system and can operate Docker or Kubernetes infrastructure. It is particularly suitable for agent developers who need tool and retrieval visibility, coding teams interested in repository impact, and organizations that must keep prompts, traces and credentials in their own environment.

It is a weaker fit for small teams that want a ready-made hosted product, formal vendor support or a published Cloud price. The OSS plan's unlimited quotas are attractive, but they do not remove the work of securing, upgrading and monitoring a self-hosted installation.

Pros and cons

Pro: One OpenTelemetry-based system covers LLM calls, tools, retrievals, evaluations, prompts and costs, reducing the need to join separate tools.

Pro: Unlimited self-hosted usage, users, projects and environments make the OSS plan workable for broad internal adoption without seat or volume charges.

Pro: Apache 2.0 licensing and local handling of prompts, traces and secrets support controlled deployments.

Con: Cloud is not yet available, so teams seeking managed hosting cannot buy that option today.

Con: Community support is the OSS support model, and Fleet Hub's collector monitoring is limited to a single cluster.

Alternatives

W&B Weave is a better fit when a hosted freemium service with a defined free allowance is preferable: its Free plan includes 1 GB per month of ingestion, 5 GB of storage, evaluations, tracing and scorers.

Opik suits teams wanting open-source observability and evaluation with a Free Cloud option, while Helicone offers an open-source, hosted-oriented route with a Hobby allowance of 10,000 requests and 1 GB of storage.

Maxim AI is more appropriate when a small managed team workspace matters: its Developer plan allows up to three seats, one workspace, 10,000 logs per month and three-day retention. Respan (formerly Keywords AI) fits buyers prioritizing explicit evaluation and prompt quotas, with 100,000 logs, 1,000 scores, five datasets, two evaluators and five prompts on Free.

Agenta is an alternative for teams wanting defined run and evaluation limits, including 5,000 agent runs and 20 evaluations per month on Hobby. LangWatch provides Apache 2.0 open-source software and a Developer allowance of 50,000 events per month with 14-day data access. Portkey is suited to teams that want a free hosted developer tier with 10,000 recorded logs per month, three-day log retention and three prompt templates.

Verdict

OpenLIT is a strong recommendation for technically capable teams that want an open-source, self-hosted control plane for LLM and agent observability, evaluation, prompts and costs. Its main reason to choose it is the unusually broad, unlimited OSS package with OpenTelemetry and local data control. Look elsewhere if managed hosting, vendor support or immediately published Cloud terms are essential.

Verdict

OpenLIT combines tracing, evaluations, prompt tools, and cost tracking with self-hosted deployment and an OSS plan listed at free. Buyers seeking a hosted option should note that the Cloud plan’s price and feature set are not yet listed.

Where Specifiction places it

Alternatives to OpenLITAll alternatives →