Curated list of open-source LLMOps tools: tracing, evals, AI gateways, guardrails and prompt management. Sorted by stars, refreshed weekly.
Python
29
1 commits
updated Oct 9, 2026
A curated list of 144 open-source tools for running LLM applications in production: tracing, evaluation, gateways, guardrails and prompt management.
Every table is sorted by GitHub stars, highest first. A GitHub Action refreshes the stars, licenses and dates every week. Last refresh: 2026-10-09.
This list covers tools you can run yourself. It leaves out papers, courses, model benchmarks and agent frameworks. Other lists cover those well.
How to read the tables. archived means the owner archived the repo on GitHub. inactive means no push in the last 12 months. † means the main code is source-available rather than under an OSI-approved open-source license, so check the terms before you use it commercially. "+ EE" means an open-source core with a separately licensed enterprise folder.
Platforms that combine at least two of tracing, evaluation, prompt management and a gateway in one product.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 35.6k | langfuse | Tracing, evaluations, prompt management and datasets for LLM apps and agents. | MIT + EE | 2026-10-09 |
| 28.3k | mlflow | ML and AI engineering platform with LLM tracing, evaluation and a prompt registry. | Apache-2.0 | 2026-10-09 |
| 22.5k | opik | Tracing, automated evaluations and production monitoring for LLM apps, RAG and agents. | Apache-2.0 | 2026-10-09 |
| 11.7k | tensorzero (archived) | LLM gateway, observability, evaluation, optimization and experimentation in one stack. | Apache-2.0 | 2026-06-11 |
| 5.8k | coze-loop | Agent development platform with prompt debugging, evaluation and trace monitoring. | Apache-2.0 | 2026-10-09 |
| 5.2k | Kiln | Desktop app and library for evals, synthetic data, fine-tuning and dataset management. | MIT (libs) + EULA (app) † | 2026-10-09 |
| 4.9k | langwatch | Agent observability, evaluations, simulations and prompt management. | Apache-2.0 | 2026-10-09 |
| 4.8k | agenta | Prompt playground, prompt versioning, evaluation and observability for LLM apps. | MIT + EE | 2026-10-09 |
| 4.7k | latitude-llm | Prompt management, evaluations and observability for AI agents. | MIT | 2026-10-09 |
| 3.3k | pezzo | Prompt design, versioning, delivery and observability for LLM apps. | Apache-2.0 | 2026-08-21 |
| 2.1k | future-agi | Tracing, evals, simulations, datasets and guardrails for LLM and agent apps. | Apache-2.0 | 2026-10-09 |
| 1.1k | weave | Tracing, evaluations and scoring for LLM apps, from Weights & Biases. | Apache-2.0 | 2026-10-09 |
| 785 | pandaprobe | Traces, evals and metrics for agents built with LangGraph, CrewAI and others. | Apache-2.0 | 2026-10-04 |
| 606 | overmind | Tracing and evaluation loop for continuously improving AI agents. | AGPL-3.0 | 2026-10-08 |
| 389 | LLMstudio | LLM proxy, prompt playground and request logging for production apps. | MPL-2.0 | 2026-07-29 |
| 343 | palico-ai (inactive) | Framework for building, experimenting with and evaluating LLM applications. | MIT | 2024-11-26 |
| 236 | starwhale (inactive) | MLOps and LLMOps platform for model evaluation, datasets and fine-tuning. | Apache-2.0 | 2024-12-20 |
| 130 | kitchenai (inactive) | LLMOps toolkit for building, testing and deploying AI app backends. | Apache-2.0 | 2025-02-14 |
| 112 | arcadia (inactive) | Kubernetes-native LLMOps platform for model serving, RAG and agent apps. | Apache-2.0 | 2024-09-21 |
| 101 | litefuse | Lightweight agent observability and evaluation platform derived from Langfuse. | MIT | 2026-10-09 |
| 11 | AcruxCore | Prompt management, an OpenAI-compatible gateway, OpenTelemetry tracing, a tool catalog and evals. | Apache-2.0 | 2026-10-09 |
Record LLM calls, agent steps, cost and latency, and let you inspect them.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 16.2k | RagaAI-Catalyst | Python SDK for agent tracing, monitoring and evaluation. | Apache-2.0 | 2026-02-11 |
| 11.8k | phoenix | Tracing, evaluation and dataset experiments built on OpenTelemetry. | Elastic-2.0 † | 2026-10-09 |
| 8.0k | evidently | Evaluation, testing and monitoring for ML models and LLM systems. | Apache-2.0 | 2026-09-29 |
| 6.2k | helicone | Proxy-based LLM observability: request logging, cost tracking, sessions and prompts. | Apache-2.0 | 2026-09-16 |
| 5.9k | agentops | Python SDK for agent monitoring, cost tracking and session replay. | MIT | 2026-06-25 |
| 4.5k | logfire | Observability for LLM and agent systems, built on OpenTelemetry, from the Pydantic team. | MIT | 2026-10-09 |
| 4.1k | deepchecks | Validation and testing for ML models and data, including LLM checks. | unknown | 2025-12-28 |
| 3.4k | lmnr | Tracing, evaluations and session replay purpose-built for AI agents. | Apache-2.0 | 2026-10-09 |
| 2.8k | openlit | OpenTelemetry-native tracing, evaluation, guardrails and GPU monitoring for AI apps. | Apache-2.0 | 2026-10-09 |
| 1.2k | langtrace | OpenTelemetry-based tracing and evaluation for LLM applications. | AGPL-3.0 | 2025-11-17 |
| 1.1k | langsmith-sdk | Client SDKs for LangSmith tracing and evaluation (the LangSmith server is not open source). | MIT | 2026-10-09 |
| 1.0k | llm.report (archived) | Logging and cost analytics for OpenAI API requests. | GPL-3.0 | 2024-05-14 |
| 998 | langkit (inactive) | Extracts quality and safety signals from prompts and responses for monitoring. | Apache-2.0 | 2024-11-22 |
| 811 | vllora | Local debugging and tracing for AI agents. | Elastic-2.0 † | 2026-06-30 |
| 795 | traceroot | Tracing and debugging layer for AI agents. | Apache-2.0 + EE | 2026-10-09 |
| 394 | agent-prism | React components for visualizing AI agent traces. | MIT | 2026-10-08 |
| 128 | agento11y | Agent observability from Grafana Labs. | Apache-2.0 | 2026-10-09 |
| 126 | agentwatch (inactive) | Observability for agent interactions across frameworks. | Apache-2.0 | 2025-05-14 |
| 126 | traccia-py | OpenTelemetry-native SDK for agent tracing, evaluation and policy enforcement. | Apache-2.0 | 2026-10-08 |
| 119 | tma1 | Local-first LLM call recorder that feeds what it sees back to the agent. | Apache-2.0 | 2026-10-08 |
| 99 | Axon | OpenTelemetry-native CLI that shows LLM and agent traces in real time. | MIT | 2026-08-06 |
| 96 | agent-trace | Shows what an agent did, what it cost and where it failed. | MIT | 2026-08-22 |
| 92 | radicalbit-ai-monitoring | Monitoring for ML models and LLMs in production. | Apache-2.0 | 2026-06-15 |
| 68 | dunetrace | Runtime reliability checks for AI agents. | Apache-2.0 | 2026-09-20 |
| 65 | prompt_sail (inactive) | LLM proxy that captures and logs every request and response. | MIT | 2025-06-15 |
| 55 | myscale-telemetry (inactive) | Tracing for LLM applications stored in MyScale. | MIT | 2025-01-02 |
Libraries that emit LLM spans in OpenTelemetry format, so any compatible backend can read them.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 7.5k | openllmetry | OpenTelemetry instrumentation for LLM providers, vector DBs and frameworks (Python). | Apache-2.0 | 2026-10-06 |
| 1.3k | openinference | OpenTelemetry semantic conventions and instrumentation for AI apps. | Apache-2.0 | 2026-10-09 |
| 412 | openllmetry-js | OpenTelemetry instrumentation for LLM apps (TypeScript). | Apache-2.0 | 2026-06-18 |
Score LLM outputs, run test suites and benchmarks, and catch regressions.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 25.8k | promptfoo | CLI and CI tool for testing prompts, agents and RAG, with red teaming. | MIT | 2026-10-09 |
| 19.6k | evals | Framework and registry of benchmarks for evaluating LLMs. | MIT | 2026-04-14 |
| 18.7k | deepeval | Pytest-style evaluation framework with LLM-as-judge metrics. | Apache-2.0 | 2026-10-07 |
| 16.0k | ragas | Evaluation metrics and test-data generation for RAG and LLM apps. | Apache-2.0 | 2026-02-24 |
| 14.2k | lm-evaluation-harness | Few-shot benchmark harness for evaluating language models. | MIT | 2026-09-14 |
| 7.5k | opencompass | Model evaluation platform across 100+ benchmark datasets. | Apache-2.0 | 2026-09-28 |
| 6.0k | harbor | Framework for evaluating and improving agents. | Apache-2.0 | 2026-10-09 |
| 5.9k | giskard-oss | Evaluation and testing library for LLM agents. | Apache-2.0 | 2026-10-09 |
| 3.6k | trulens | Evaluation and tracking for LLM experiments and agents. | MIT | 2026-10-09 |
| 3.1k | prompttools | Tools for prompt testing and experiments across models and vector DBs. | Apache-2.0 | 2026-02-11 |
| 3.0k | ChainForge | Visual environment for comparing prompts and models side by side. | MIT | 2026-10-03 |
| 3.0k | inspect_ai | Framework for LLM evaluations from the UK AI Security Institute. | MIT | 2026-10-09 |
| 2.9k | helm | Holistic Evaluation of Language Models, from Stanford CRFM. | Apache-2.0 | 2026-09-01 |
| 2.4k | uptrain (inactive) | Preconfigured checks for grading LLM responses. | Apache-2.0 | 2024-08-18 |
| 2.2k | inspector | Testing and evaluation tool for MCP servers and apps. | Apache-2.0 + commercial | 2026-10-09 |
| 1.7k | evalite | TypeScript eval runner with a local UI. | MIT | 2026-04-28 |
| 1.3k | intellagent | Agent evaluation through simulated, synthetic conversations. | Apache-2.0 | 2026-09-14 |
| 1.2k | openevals | Ready-made evaluators for LLM apps. | MIT | 2026-10-02 |
| 1.2k | uqlm | Hallucination detection using uncertainty quantification. | Apache-2.0 | 2026-10-08 |
| 1.1k | prometheus-eval (inactive) | Open evaluator models for scoring LLM responses. | Apache-2.0 | 2025-04-25 |
| 1.1k | judgeval | Evals and monitoring for agents. | Apache-2.0 | 2026-10-07 |
| 1.1k | autoevals | Library of evaluators for scoring AI model outputs. | MIT | 2026-10-08 |
| 519 | continuous-eval | Data-driven, modular evaluation for LLM pipelines. | Apache-2.0 | 2026-08-10 |
| 397 | rhesis | Test generation and review workflows for LLM apps and agents. | MIT + EE | 2026-10-08 |
| 328 | tonic_validate (inactive) | Metrics for evaluating RAG responses. | MIT | 2025-07-10 |
| 166 | empirical (inactive) | Test and compare LLMs and model configurations. | MIT | 2024-08-20 |
| 164 | promptpex | Generates test cases from prompts. | CC-BY-4.0 | 2026-10-03 |
| 163 | agentevals | Framework-agnostic agent evaluations from OpenTelemetry traces. | Apache-2.0 | 2026-10-09 |
| 161 | pytest-evals (inactive) | Pytest plugin for LLM evaluation tests. | MIT | 2025-02-05 |
| 108 | autoarena (inactive) | Ranks models, RAG systems and prompts through head-to-head judging. | Apache-2.0 | 2024-12-16 |
| 60 | dokimos | LLM and agent evaluation for Java and Kotlin, run in JUnit. | MIT | 2026-10-07 |
| 52 | viteval | LLM evaluation framework built on Vitest. | MIT | 2026-05-23 |
| 52 | litmus | LLM testing and evaluation tool with a web UI. | Apache-2.0 | 2026-03-29 |
Attack your own LLM app before someone else does: jailbreaks, prompt injection, data leaks.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 9.5k | garak | LLM vulnerability scanner. | Apache-2.0 | 2026-10-09 |
| 6.8k | AI-Infra-Guard | Red-teaming platform covering agents, MCP servers and jailbreaks. | Apache-2.0 | 2026-10-09 |
| 3.0k | deepteam | Framework for red teaming LLMs and agents. | Apache-2.0 | 2026-10-01 |
| 2.0k | agentic_security | Vulnerability scanner and red-teaming kit for LLM apps. | Apache-2.0 | 2026-09-22 |
| 1.6k | FuzzyAI | Automated fuzzing to find jailbreaks in LLMs. | Apache-2.0 | 2026-02-06 |
Check inputs and outputs at runtime and block what should not pass.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 7.5k | guardrails | Validators that check and correct LLM inputs and outputs. | Apache-2.0 | 2026-10-09 |
| 7.3k | Guardrails | Programmable guardrails for conversational LLM systems. | Apache-2.0 | 2026-10-09 |
| 3.2k | llm-guard (archived) | Input and output scanners for LLM interactions. | MIT | 2026-07-08 |
| 1.5k | rebuff (archived) | Prompt-injection detector. | Apache-2.0 | 2024-08-07 |
| 505 | vigil-llm (inactive) | Detects prompt injections and jailbreaks in LLM inputs. | Apache-2.0 | 2024-01-31 |
| 181 | gateway (inactive) | Self-hosted LLM firewall with configurable guardrails. | BUSL-1.1 † | 2025-06-25 |
| 155 | fast-llm-security-guardrails | Guardrails for prompt injection, PII and topic control. | MIT | 2026-02-03 |
| 136 | last_layer (inactive) | Low-latency prompt-injection and jailbreak detection. | MIT | 2024-07-26 |
| 52 | hai-guardrails | TypeScript guards for LLM applications. | MIT | 2026-02-06 |
One API in front of many model providers, with routing, fallbacks, keys, budgets and logs.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 74.6k | OmniRoute | OpenAI-compatible gateway over hundreds of providers, many with free tiers. | MIT | 2026-10-09 |
| 60.6k | litellm | Python SDK and proxy server for 100+ LLM APIs, with cost tracking and budgets. | MIT + EE | 2026-10-09 |
| 49.5k | new-api | Model aggregation gateway that converts between OpenAI, Claude and Gemini formats. | AGPL-3.0 | 2026-10-09 |
| 37.1k | one-api | LLM API key management and distribution behind one OpenAI-style API. | MIT | 2026-01-09 |
| 13.2k | gateway | AI gateway with routing, retries, caching and built-in guardrails. | MIT | 2026-05-25 |
| 8.7k | bifrost | Go AI gateway with load balancing, clustering, guardrails and budgets. | Apache-2.0 | 2026-10-09 |
| 7.6k | llm-gateway | Self-hosted LLM gateway from Manifest. | MIT | 2026-10-09 |
| 7.1k | plano | Proxy and data plane for agent apps: model routing, observability and guardrails. | Apache-2.0 | 2026-10-07 |
| 7.1k | gpt-load | Self-hosted gateway that pools keys and accounts, with failover and request logs. | MIT | 2026-10-09 |
| 6.1k | semantic-router | Routes each request to a model based on its content, from the vLLM project. | Apache-2.0 | 2026-10-09 |
| 5.3k | axonhub | AI gateway with failover, load balancing, cost control and tracing. | Apache-2.0 | 2026-10-08 |
| 2.2k | agent-router | AI gateway built on Envoy Gateway (formerly Envoy AI Gateway). | Apache-2.0 | 2026-10-09 |
| 1.7k | llmgateway | Routes, manages and analyzes LLM requests across providers. | AGPL-3.0 + EE | 2026-10-09 |
| 1.2k | GoModel | Go AI gateway with OpenAI- and Anthropic-compatible APIs. | MIT | 2026-10-09 |
| 818 | ThinkWatch | Self-hosted AI and MCP gateway with SSO, RBAC and PII redaction. | BUSL-1.1 † | 2026-10-09 |
| 657 | NadirClaw | Routes simple prompts to cheap models and hard ones to premium models. | PolyForm-Noncommercial-1.0.0 † | 2026-10-06 |
| 638 | ai-gateway | Rust AI gateway from Helicone. | GPL-3.0 | 2025-11-21 |
| 607 | gateway | Local SDK with one interface to 200+ LLMs. | MIT | 2026-07-29 |
| 564 | smg | Rust LLM gateway across self-hosted engines and hosted APIs. | Apache-2.0 | 2026-10-09 |
| 561 | aiproxy | AI gateway with multi-channel routing and error handling. | MIT | 2026-09-23 |
| 556 | LLM-API-Key-Proxy | OpenAI- and Anthropic-compatible proxy with key rotation and load balancing. | LGPL-3.0 / MIT | 2026-09-23 |
| 515 | otari | Self-hosted OpenAI-compatible gateway with virtual keys and budgets, from Mozilla.ai. | Apache-2.0 | 2026-10-09 |
| 454 | GPTRouter (inactive) | Routes across LLM and image models with fallbacks. | MIT | 2024-04-10 |
| 318 | llmio | LLM gateway with weighted load balancing and cost tracking. | MIT | 2026-10-06 |
| 264 | ai-gateway | Go AI gateway with caching, guardrails, A/B tests and cost controls. | Apache-2.0 | 2026-10-04 |
| 230 | hub | Rust LLM gateway with OpenTelemetry tracing. | Apache-2.0 | 2026-07-16 |
| 215 | yolorouter | Self-hosted gateway with failover, key rotation and an admin console. | Apache-2.0 | 2026-10-03 |
| 182 | aisix | Rust AI gateway with routing and guardrails, from API7. | Apache-2.0 | 2026-10-09 |
| 160 | glide (inactive) | Model gateway for production GenAI apps. | Apache-2.0 | 2024-08-12 |
| 99 | llm-gateway | Zero-trust LLM gateway with semantic routing. | Apache-2.0 | 2026-09-15 |
General API gateways that added LLM features, and gateways for MCP and agent traffic.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 44.3k | kong | API gateway with AI plugins for multi-LLM routing and governance. | Apache-2.0 | 2026-10-09 |
| 17.2k | apisix | Cloud-native API gateway with AI proxy plugins. | Apache-2.0 | 2026-10-09 |
| 10.9k | tyk | API gateway with AI and MCP support. | MPL-2.0 + EE | 2026-10-09 |
| 9.5k | higress | AI-native API gateway built on Envoy and Istio. | Apache-2.0 | 2026-10-08 |
| 5.7k | kgateway | Kubernetes-native API gateway with AI gateway features. | Apache-2.0 | 2026-10-09 |
| 5.3k | agentgateway | Proxy for agent-to-agent and MCP traffic. | Apache-2.0 | 2026-10-09 |
| 4.6k | mcp-context-forge | Gateway and registry in front of MCP, A2A and REST services. | Apache-2.0 | 2026-10-09 |
| 1.8k | APIPark | AI and API gateway with an API developer portal. | Apache-2.0 | 2026-09-09 |
Store, version, test and serve prompts outside your application code.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 4.6k | Promptify | Prompt templates and versioning for structured LLM output. | Apache-2.0 | 2026-03-27 |
| 1.3k | prompty | File format and tooling for creating, running and evaluating prompts. | MIT | 2026-10-05 |
| 1.1k | aiconfig | Config-based format for storing prompts and model settings. | MIT | 2026-02-10 |
| 268 | PromptWorks | Prompt management and testing tool. | Apache-2.0 | 2026-06-22 |
| 181 | prompt-shelf (inactive) | Lightweight prompt management system. | MIT | 2025-07-30 |
| 121 | llmkit (inactive) | Prompt versioning, testing and evaluation server with a UI. | MIT | 2025-06-26 |
| 112 | deck | Laravel package for versioned prompts with A/B testing. | MIT | 2026-10-09 |
| 108 | proofhound | Prompt management, optimization and release platform. | Apache-2.0 | 2026-07-03 |
| 99 | promptdesk (inactive) | Create, organize and evaluate prompts. | AGPL-3.0 | 2025-08-29 |
Additions and corrections are welcome. Read CONTRIBUTING.md first. In short: edit data/projects.toml, not this README.
This list is maintained by Talha Anwar, who also builds AcruxCore, one of the platforms listed above. AcruxCore is sorted by stars like every other entry.
CC0 1.0. To the extent possible under law, the maintainer has waived all copyright to this list.
Curated list of open-source LLMOps tools: tracing, evals, AI gateways, guardrails and prompt management. Sorted by stars, refreshed weekly.
Python
29
1 commits
updated Oct 9, 2026
A curated list of 144 open-source tools for running LLM applications in production: tracing, evaluation, gateways, guardrails and prompt management.
Every table is sorted by GitHub stars, highest first. A GitHub Action refreshes the stars, licenses and dates every week. Last refresh: 2026-10-09.
This list covers tools you can run yourself. It leaves out papers, courses, model benchmarks and agent frameworks. Other lists cover those well.
How to read the tables. archived means the owner archived the repo on GitHub. inactive means no push in the last 12 months. † means the main code is source-available rather than under an OSI-approved open-source license, so check the terms before you use it commercially. "+ EE" means an open-source core with a separately licensed enterprise folder.
Platforms that combine at least two of tracing, evaluation, prompt management and a gateway in one product.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 35.6k | langfuse | Tracing, evaluations, prompt management and datasets for LLM apps and agents. | MIT + EE | 2026-10-09 |
| 28.3k | mlflow | ML and AI engineering platform with LLM tracing, evaluation and a prompt registry. | Apache-2.0 | 2026-10-09 |
| 22.5k | opik | Tracing, automated evaluations and production monitoring for LLM apps, RAG and agents. | Apache-2.0 | 2026-10-09 |
| 11.7k | tensorzero (archived) | LLM gateway, observability, evaluation, optimization and experimentation in one stack. | Apache-2.0 | 2026-06-11 |
| 5.8k | coze-loop | Agent development platform with prompt debugging, evaluation and trace monitoring. | Apache-2.0 | 2026-10-09 |
| 5.2k | Kiln | Desktop app and library for evals, synthetic data, fine-tuning and dataset management. | MIT (libs) + EULA (app) † | 2026-10-09 |
| 4.9k | langwatch | Agent observability, evaluations, simulations and prompt management. | Apache-2.0 | 2026-10-09 |
| 4.8k | agenta | Prompt playground, prompt versioning, evaluation and observability for LLM apps. | MIT + EE | 2026-10-09 |
| 4.7k | latitude-llm | Prompt management, evaluations and observability for AI agents. | MIT | 2026-10-09 |
| 3.3k | pezzo | Prompt design, versioning, delivery and observability for LLM apps. | Apache-2.0 | 2026-08-21 |
| 2.1k | future-agi | Tracing, evals, simulations, datasets and guardrails for LLM and agent apps. | Apache-2.0 | 2026-10-09 |
| 1.1k | weave | Tracing, evaluations and scoring for LLM apps, from Weights & Biases. | Apache-2.0 | 2026-10-09 |
| 785 | pandaprobe | Traces, evals and metrics for agents built with LangGraph, CrewAI and others. | Apache-2.0 | 2026-10-04 |
| 606 | overmind | Tracing and evaluation loop for continuously improving AI agents. | AGPL-3.0 | 2026-10-08 |
| 389 | LLMstudio | LLM proxy, prompt playground and request logging for production apps. | MPL-2.0 | 2026-07-29 |
| 343 | palico-ai (inactive) | Framework for building, experimenting with and evaluating LLM applications. | MIT | 2024-11-26 |
| 236 | starwhale (inactive) | MLOps and LLMOps platform for model evaluation, datasets and fine-tuning. | Apache-2.0 | 2024-12-20 |
| 130 | kitchenai (inactive) | LLMOps toolkit for building, testing and deploying AI app backends. | Apache-2.0 | 2025-02-14 |
| 112 | arcadia (inactive) | Kubernetes-native LLMOps platform for model serving, RAG and agent apps. | Apache-2.0 | 2024-09-21 |
| 101 | litefuse | Lightweight agent observability and evaluation platform derived from Langfuse. | MIT | 2026-10-09 |
| 11 | AcruxCore | Prompt management, an OpenAI-compatible gateway, OpenTelemetry tracing, a tool catalog and evals. | Apache-2.0 | 2026-10-09 |
Record LLM calls, agent steps, cost and latency, and let you inspect them.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 16.2k | RagaAI-Catalyst | Python SDK for agent tracing, monitoring and evaluation. | Apache-2.0 | 2026-02-11 |
| 11.8k | phoenix | Tracing, evaluation and dataset experiments built on OpenTelemetry. | Elastic-2.0 † | 2026-10-09 |
| 8.0k | evidently | Evaluation, testing and monitoring for ML models and LLM systems. | Apache-2.0 | 2026-09-29 |
| 6.2k | helicone | Proxy-based LLM observability: request logging, cost tracking, sessions and prompts. | Apache-2.0 | 2026-09-16 |
| 5.9k | agentops | Python SDK for agent monitoring, cost tracking and session replay. | MIT | 2026-06-25 |
| 4.5k | logfire | Observability for LLM and agent systems, built on OpenTelemetry, from the Pydantic team. | MIT | 2026-10-09 |
| 4.1k | deepchecks | Validation and testing for ML models and data, including LLM checks. | unknown | 2025-12-28 |
| 3.4k | lmnr | Tracing, evaluations and session replay purpose-built for AI agents. | Apache-2.0 | 2026-10-09 |
| 2.8k | openlit | OpenTelemetry-native tracing, evaluation, guardrails and GPU monitoring for AI apps. | Apache-2.0 | 2026-10-09 |
| 1.2k | langtrace | OpenTelemetry-based tracing and evaluation for LLM applications. | AGPL-3.0 | 2025-11-17 |
| 1.1k | langsmith-sdk | Client SDKs for LangSmith tracing and evaluation (the LangSmith server is not open source). | MIT | 2026-10-09 |
| 1.0k | llm.report (archived) | Logging and cost analytics for OpenAI API requests. | GPL-3.0 | 2024-05-14 |
| 998 | langkit (inactive) | Extracts quality and safety signals from prompts and responses for monitoring. | Apache-2.0 | 2024-11-22 |
| 811 | vllora | Local debugging and tracing for AI agents. | Elastic-2.0 † | 2026-06-30 |
| 795 | traceroot | Tracing and debugging layer for AI agents. | Apache-2.0 + EE | 2026-10-09 |
| 394 | agent-prism | React components for visualizing AI agent traces. | MIT | 2026-10-08 |
| 128 | agento11y | Agent observability from Grafana Labs. | Apache-2.0 | 2026-10-09 |
| 126 | agentwatch (inactive) | Observability for agent interactions across frameworks. | Apache-2.0 | 2025-05-14 |
| 126 | traccia-py | OpenTelemetry-native SDK for agent tracing, evaluation and policy enforcement. | Apache-2.0 | 2026-10-08 |
| 119 | tma1 | Local-first LLM call recorder that feeds what it sees back to the agent. | Apache-2.0 | 2026-10-08 |
| 99 | Axon | OpenTelemetry-native CLI that shows LLM and agent traces in real time. | MIT | 2026-08-06 |
| 96 | agent-trace | Shows what an agent did, what it cost and where it failed. | MIT | 2026-08-22 |
| 92 | radicalbit-ai-monitoring | Monitoring for ML models and LLMs in production. | Apache-2.0 | 2026-06-15 |
| 68 | dunetrace | Runtime reliability checks for AI agents. | Apache-2.0 | 2026-09-20 |
| 65 | prompt_sail (inactive) | LLM proxy that captures and logs every request and response. | MIT | 2025-06-15 |
| 55 | myscale-telemetry (inactive) | Tracing for LLM applications stored in MyScale. | MIT | 2025-01-02 |
Libraries that emit LLM spans in OpenTelemetry format, so any compatible backend can read them.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 7.5k | openllmetry | OpenTelemetry instrumentation for LLM providers, vector DBs and frameworks (Python). | Apache-2.0 | 2026-10-06 |
| 1.3k | openinference | OpenTelemetry semantic conventions and instrumentation for AI apps. | Apache-2.0 | 2026-10-09 |
| 412 | openllmetry-js | OpenTelemetry instrumentation for LLM apps (TypeScript). | Apache-2.0 | 2026-06-18 |
Score LLM outputs, run test suites and benchmarks, and catch regressions.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 25.8k | promptfoo | CLI and CI tool for testing prompts, agents and RAG, with red teaming. | MIT | 2026-10-09 |
| 19.6k | evals | Framework and registry of benchmarks for evaluating LLMs. | MIT | 2026-04-14 |
| 18.7k | deepeval | Pytest-style evaluation framework with LLM-as-judge metrics. | Apache-2.0 | 2026-10-07 |
| 16.0k | ragas | Evaluation metrics and test-data generation for RAG and LLM apps. | Apache-2.0 | 2026-02-24 |
| 14.2k | lm-evaluation-harness | Few-shot benchmark harness for evaluating language models. | MIT | 2026-09-14 |
| 7.5k | opencompass | Model evaluation platform across 100+ benchmark datasets. | Apache-2.0 | 2026-09-28 |
| 6.0k | harbor | Framework for evaluating and improving agents. | Apache-2.0 | 2026-10-09 |
| 5.9k | giskard-oss | Evaluation and testing library for LLM agents. | Apache-2.0 | 2026-10-09 |
| 3.6k | trulens | Evaluation and tracking for LLM experiments and agents. | MIT | 2026-10-09 |
| 3.1k | prompttools | Tools for prompt testing and experiments across models and vector DBs. | Apache-2.0 | 2026-02-11 |
| 3.0k | ChainForge | Visual environment for comparing prompts and models side by side. | MIT | 2026-10-03 |
| 3.0k | inspect_ai | Framework for LLM evaluations from the UK AI Security Institute. | MIT | 2026-10-09 |
| 2.9k | helm | Holistic Evaluation of Language Models, from Stanford CRFM. | Apache-2.0 | 2026-09-01 |
| 2.4k | uptrain (inactive) | Preconfigured checks for grading LLM responses. | Apache-2.0 | 2024-08-18 |
| 2.2k | inspector | Testing and evaluation tool for MCP servers and apps. | Apache-2.0 + commercial | 2026-10-09 |
| 1.7k | evalite | TypeScript eval runner with a local UI. | MIT | 2026-04-28 |
| 1.3k | intellagent | Agent evaluation through simulated, synthetic conversations. | Apache-2.0 | 2026-09-14 |
| 1.2k | openevals | Ready-made evaluators for LLM apps. | MIT | 2026-10-02 |
| 1.2k | uqlm | Hallucination detection using uncertainty quantification. | Apache-2.0 | 2026-10-08 |
| 1.1k | prometheus-eval (inactive) | Open evaluator models for scoring LLM responses. | Apache-2.0 | 2025-04-25 |
| 1.1k | judgeval | Evals and monitoring for agents. | Apache-2.0 | 2026-10-07 |
| 1.1k | autoevals | Library of evaluators for scoring AI model outputs. | MIT | 2026-10-08 |
| 519 | continuous-eval | Data-driven, modular evaluation for LLM pipelines. | Apache-2.0 | 2026-08-10 |
| 397 | rhesis | Test generation and review workflows for LLM apps and agents. | MIT + EE | 2026-10-08 |
| 328 | tonic_validate (inactive) | Metrics for evaluating RAG responses. | MIT | 2025-07-10 |
| 166 | empirical (inactive) | Test and compare LLMs and model configurations. | MIT | 2024-08-20 |
| 164 | promptpex | Generates test cases from prompts. | CC-BY-4.0 | 2026-10-03 |
| 163 | agentevals | Framework-agnostic agent evaluations from OpenTelemetry traces. | Apache-2.0 | 2026-10-09 |
| 161 | pytest-evals (inactive) | Pytest plugin for LLM evaluation tests. | MIT | 2025-02-05 |
| 108 | autoarena (inactive) | Ranks models, RAG systems and prompts through head-to-head judging. | Apache-2.0 | 2024-12-16 |
| 60 | dokimos | LLM and agent evaluation for Java and Kotlin, run in JUnit. | MIT | 2026-10-07 |
| 52 | viteval | LLM evaluation framework built on Vitest. | MIT | 2026-05-23 |
| 52 | litmus | LLM testing and evaluation tool with a web UI. | Apache-2.0 | 2026-03-29 |
Attack your own LLM app before someone else does: jailbreaks, prompt injection, data leaks.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 9.5k | garak | LLM vulnerability scanner. | Apache-2.0 | 2026-10-09 |
| 6.8k | AI-Infra-Guard | Red-teaming platform covering agents, MCP servers and jailbreaks. | Apache-2.0 | 2026-10-09 |
| 3.0k | deepteam | Framework for red teaming LLMs and agents. | Apache-2.0 | 2026-10-01 |
| 2.0k | agentic_security | Vulnerability scanner and red-teaming kit for LLM apps. | Apache-2.0 | 2026-09-22 |
| 1.6k | FuzzyAI | Automated fuzzing to find jailbreaks in LLMs. | Apache-2.0 | 2026-02-06 |
Check inputs and outputs at runtime and block what should not pass.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 7.5k | guardrails | Validators that check and correct LLM inputs and outputs. | Apache-2.0 | 2026-10-09 |
| 7.3k | Guardrails | Programmable guardrails for conversational LLM systems. | Apache-2.0 | 2026-10-09 |
| 3.2k | llm-guard (archived) | Input and output scanners for LLM interactions. | MIT | 2026-07-08 |
| 1.5k | rebuff (archived) | Prompt-injection detector. | Apache-2.0 | 2024-08-07 |
| 505 | vigil-llm (inactive) | Detects prompt injections and jailbreaks in LLM inputs. | Apache-2.0 | 2024-01-31 |
| 181 | gateway (inactive) | Self-hosted LLM firewall with configurable guardrails. | BUSL-1.1 † | 2025-06-25 |
| 155 | fast-llm-security-guardrails | Guardrails for prompt injection, PII and topic control. | MIT | 2026-02-03 |
| 136 | last_layer (inactive) | Low-latency prompt-injection and jailbreak detection. | MIT | 2024-07-26 |
| 52 | hai-guardrails | TypeScript guards for LLM applications. | MIT | 2026-02-06 |
One API in front of many model providers, with routing, fallbacks, keys, budgets and logs.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 74.6k | OmniRoute | OpenAI-compatible gateway over hundreds of providers, many with free tiers. | MIT | 2026-10-09 |
| 60.6k | litellm | Python SDK and proxy server for 100+ LLM APIs, with cost tracking and budgets. | MIT + EE | 2026-10-09 |
| 49.5k | new-api | Model aggregation gateway that converts between OpenAI, Claude and Gemini formats. | AGPL-3.0 | 2026-10-09 |
| 37.1k | one-api | LLM API key management and distribution behind one OpenAI-style API. | MIT | 2026-01-09 |
| 13.2k | gateway | AI gateway with routing, retries, caching and built-in guardrails. | MIT | 2026-05-25 |
| 8.7k | bifrost | Go AI gateway with load balancing, clustering, guardrails and budgets. | Apache-2.0 | 2026-10-09 |
| 7.6k | llm-gateway | Self-hosted LLM gateway from Manifest. | MIT | 2026-10-09 |
| 7.1k | plano | Proxy and data plane for agent apps: model routing, observability and guardrails. | Apache-2.0 | 2026-10-07 |
| 7.1k | gpt-load | Self-hosted gateway that pools keys and accounts, with failover and request logs. | MIT | 2026-10-09 |
| 6.1k | semantic-router | Routes each request to a model based on its content, from the vLLM project. | Apache-2.0 | 2026-10-09 |
| 5.3k | axonhub | AI gateway with failover, load balancing, cost control and tracing. | Apache-2.0 | 2026-10-08 |
| 2.2k | agent-router | AI gateway built on Envoy Gateway (formerly Envoy AI Gateway). | Apache-2.0 | 2026-10-09 |
| 1.7k | llmgateway | Routes, manages and analyzes LLM requests across providers. | AGPL-3.0 + EE | 2026-10-09 |
| 1.2k | GoModel | Go AI gateway with OpenAI- and Anthropic-compatible APIs. | MIT | 2026-10-09 |
| 818 | ThinkWatch | Self-hosted AI and MCP gateway with SSO, RBAC and PII redaction. | BUSL-1.1 † | 2026-10-09 |
| 657 | NadirClaw | Routes simple prompts to cheap models and hard ones to premium models. | PolyForm-Noncommercial-1.0.0 † | 2026-10-06 |
| 638 | ai-gateway | Rust AI gateway from Helicone. | GPL-3.0 | 2025-11-21 |
| 607 | gateway | Local SDK with one interface to 200+ LLMs. | MIT | 2026-07-29 |
| 564 | smg | Rust LLM gateway across self-hosted engines and hosted APIs. | Apache-2.0 | 2026-10-09 |
| 561 | aiproxy | AI gateway with multi-channel routing and error handling. | MIT | 2026-09-23 |
| 556 | LLM-API-Key-Proxy | OpenAI- and Anthropic-compatible proxy with key rotation and load balancing. | LGPL-3.0 / MIT | 2026-09-23 |
| 515 | otari | Self-hosted OpenAI-compatible gateway with virtual keys and budgets, from Mozilla.ai. | Apache-2.0 | 2026-10-09 |
| 454 | GPTRouter (inactive) | Routes across LLM and image models with fallbacks. | MIT | 2024-04-10 |
| 318 | llmio | LLM gateway with weighted load balancing and cost tracking. | MIT | 2026-10-06 |
| 264 | ai-gateway | Go AI gateway with caching, guardrails, A/B tests and cost controls. | Apache-2.0 | 2026-10-04 |
| 230 | hub | Rust LLM gateway with OpenTelemetry tracing. | Apache-2.0 | 2026-07-16 |
| 215 | yolorouter | Self-hosted gateway with failover, key rotation and an admin console. | Apache-2.0 | 2026-10-03 |
| 182 | aisix | Rust AI gateway with routing and guardrails, from API7. | Apache-2.0 | 2026-10-09 |
| 160 | glide (inactive) | Model gateway for production GenAI apps. | Apache-2.0 | 2024-08-12 |
| 99 | llm-gateway | Zero-trust LLM gateway with semantic routing. | Apache-2.0 | 2026-09-15 |
General API gateways that added LLM features, and gateways for MCP and agent traffic.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 44.3k | kong | API gateway with AI plugins for multi-LLM routing and governance. | Apache-2.0 | 2026-10-09 |
| 17.2k | apisix | Cloud-native API gateway with AI proxy plugins. | Apache-2.0 | 2026-10-09 |
| 10.9k | tyk | API gateway with AI and MCP support. | MPL-2.0 + EE | 2026-10-09 |
| 9.5k | higress | AI-native API gateway built on Envoy and Istio. | Apache-2.0 | 2026-10-08 |
| 5.7k | kgateway | Kubernetes-native API gateway with AI gateway features. | Apache-2.0 | 2026-10-09 |
| 5.3k | agentgateway | Proxy for agent-to-agent and MCP traffic. | Apache-2.0 | 2026-10-09 |
| 4.6k | mcp-context-forge | Gateway and registry in front of MCP, A2A and REST services. | Apache-2.0 | 2026-10-09 |
| 1.8k | APIPark | AI and API gateway with an API developer portal. | Apache-2.0 | 2026-09-09 |
Store, version, test and serve prompts outside your application code.
| Stars | Project | Description | License | Last push |
|---|---|---|---|---|
| 4.6k | Promptify | Prompt templates and versioning for structured LLM output. | Apache-2.0 | 2026-03-27 |
| 1.3k | prompty | File format and tooling for creating, running and evaluating prompts. | MIT | 2026-10-05 |
| 1.1k | aiconfig | Config-based format for storing prompts and model settings. | MIT | 2026-02-10 |
| 268 | PromptWorks | Prompt management and testing tool. | Apache-2.0 | 2026-06-22 |
| 181 | prompt-shelf (inactive) | Lightweight prompt management system. | MIT | 2025-07-30 |
| 121 | llmkit (inactive) | Prompt versioning, testing and evaluation server with a UI. | MIT | 2025-06-26 |
| 112 | deck | Laravel package for versioned prompts with A/B testing. | MIT | 2026-10-09 |
| 108 | proofhound | Prompt management, optimization and release platform. | Apache-2.0 | 2026-07-03 |
| 99 | promptdesk (inactive) | Create, organize and evaluate prompts. | AGPL-3.0 | 2025-08-29 |
Additions and corrections are welcome. Read CONTRIBUTING.md first. In short: edit data/projects.toml, not this README.
This list is maintained by Talha Anwar, who also builds AcruxCore, one of the platforms listed above. AcruxCore is sorted by stars like every other entry.
CC0 1.0. To the extent possible under law, the maintainer has waived all copyright to this list.
