Product / UX / Telemetry & observability

RagaAI Catalyst

Python client for RagaAI's hosted LLM observability and evaluation platform, providing project, dataset and prompt management, agentic trace recording, metric-based evaluation, synthetic data generation and guardrail management. Every operation authenticates against the hosted catalyst.raga.ai service; the repository ships no server, dashboard or self-host path of its own.

The Apache-2.0 license covers only this client library. The repository's own README requires an access key and secret key issued by the hosted catalyst.raga.ai platform for every operation and contains no docker-compose, server or dashboard code, so the product it talks to is a closed SaaS. No artifact is declared, on purpose. The scored product is the hosted platform, and both the raga-ai-hub/RagaAI-Catalyst repository and the ragaai-catalyst PyPI package are its client, so installing or starring them measures the client, not the platform - the open-satellite-around-a-closed-core shape in docs/reference/identity.md, which is why the repository is recorded in the resolution ledger against this product rather than declared. Both stay cited as openness evidence, because that is what they are evidence of. Verified 2026-09-02 via the raga-ai-hub/RagaAI-Catalyst repository.

Openness

1 high confidence

1.0

license: Apache-2.0(covers only the ragaai-catalyst client library, not the platform it talks to)
source: closed(the observability/eval platform (project, dataset, trace and prompt storage, evaluation execution) runs at catalyst.raga.ai
platform: proprietary-SaaS(every class in the README requires an access_key/secret_key issued by the hosted platform)
self-host: no(no server, dashboard or docker-compose file anywhere in the repository)

The repository is a well-featured client SDK - project, dataset, evaluation, trace, prompt, synthetic-data and guardrail management - but every one of those calls authenticates against catalyst.raga.ai and none of the platform it talks to is published. A full recursive tree search found no docker, compose, server, backend or dashboard path, so there is no self-host route - this is the LangSmith/PromptLayer shape, an open client over a closed hosted product, not the Helicone/Opik shape where the whole stack ships.

https://raw.githubusercontent.com/raga-ai-hub/RagaAI-Catalyst/main/LICENSE recorded 2026-09-02
Apache License, Version 2.0 full text, the sole license file in the repository. It covers this client library, not the hosted platform it authenticates against.
https://raw.githubusercontent.com/raga-ai-hub/RagaAI-Catalyst/main/README.md recorded 2026-09-02
The Configuration section: "Before using RagaAI Catalyst, you need to set up your credentials" via `access_key`, `secret_key` and `base_url`, with the note "Authentication to RagaAICatalyst is necessary to perform any operations below." Every subsequent usage example (Project, Dataset, Evaluation, Trace and Prompt Management) instantiates against that authenticated client.
https://api.github.com/repos/raga-ai-hub/RagaAI-Catalyst/git/trees/main?recursive=1 recorded 2026-09-02
The whole repository tree, 284 paths, truncated:false. Searched for docker, compose, self-host, server, backend and dashboard paths: none exist. The only license-related path is the root LICENSE file (Apache-2.0).

Adoption

not assessed

No instrument measures the hosted platform, so no level is recorded. The ragaai-catalyst PyPI package (2,711 downloads a month) is the client SDK, and a client's installs count the client's users rather than the platform's; the repository's 16,157 stars are the same client's, and stars three orders above a package's monthly downloads is the magnitude tell the adoption guide names, not a usage reading. The vendor side offers nothing to read either - raga.ai serves a JavaScript shell whose only text is a healthcare-agents pitch with no customer roster or count for Catalyst, and catalyst.raga.ai, the platform endpoint every SDK call authenticates against, answers 404. Banding the platform off any of that would be inventing the number.

https://pypistats.org/api/packages/ragaai-catalyst/recent recorded 2026-09-02
last_month = 2,711 downloads for ragaai-catalyst (last_week 359, last_day 236). A measurement of the client SDK's channel, recorded as context and explicitly not the basis of a band.
https://api.github.com/repos/raga-ai-hub/RagaAI-Catalyst recorded 2026-09-02
stargazers_count = 16,157, forks_count = 3,569, pushed_at 2026-02-11, for the client SDK repository.
https://raga.ai/ recorded 2026-09-02
A 3 KB HTML shell for a client-rendered app. The only readable text is the title and meta description - "RagaAI | Enterprise Healthcare AI Agent Suites", "purpose-built, production-grade AI Agent Suites for Healthcare, Lifesciences, and Aerospace. Powered by Prism and Catalyst" - with no customer roster, user count or download figure for Catalyst anywhere in the served body.

Capability

4 medium confidence

4.0

A broad feature matrix covering every dimension this category grades - tracing, evaluation, prompt and dataset management - plus synthetic data generation, guardrails and red-teaming that few peers in this category offer. That breadth puts it level with the Langfuse anchor at 4, even though (unlike Langfuse) none of it is self-hostable.

https://raw.githubusercontent.com/raga-ai-hub/RagaAI-Catalyst/main/README.md recorded 2026-09-02
Table of contents and usage sections for Project Management, Dataset Management, Evaluation, Trace Management, Agentic Tracing, Prompt Management, Synthetic Data Generation, Guardrail Management and Red-teaming, each with a working code example.

Verified 2026-09-02