Agenta

license

MIT(OSI, core)

source

public

commercial

Agenta-Cloud-managed-tier(free tier available)

core-gated

gated(the root LICENSE carves every ee/ directory out of MIT and under ee/LICENSE, the Agenta Enterprise License, which permits production use only with a valid Agenta Enterprise License

MIT core with a commercially licensed enterprise half in the same repository, self-hostable either way. The gating is read directly from the repository and does not rest on Agenta Cloud existing. The root LICENSE splits the repo: "All content that resides under any 'ee/' directory of this repository, if such directories exist, are licensed under the license defined in 'ee/LICENSE'", and everything else is MIT Expat. ee/LICENSE is the Agenta Enterprise License, under which the software "may only be used in production, if you (and any entity that you represent) have agreed to, and are in compliance with, the Agenta Subscription Terms of Service ... and otherwise have a valid Agenta Enterprise License"; copying and modifying for development and testing is allowed without a subscription, and "it is forbidden to copy, merge, publish, distribute, sublicense, and/or sell the Software". The carve-out is real code rather than a stub: the api tree splits into api/oss and api/ee, and web/ee and services/ee exist alongside. Self-hostable and gated are both true of the same repository - the MIT half genuinely runs on your own hardware, and the enterprise half is the part you may not run in production without paying.

• https://github.com/Agenta-AI/agenta recorded 2026-06-04

  MIT license; self-host + Agenta Cloud managed tier

• https://raw.githubusercontent.com/Agenta-AI/agenta/main/LICENSE recorded 2026-08-12

  Root LICENSE. "Portions of this software are licensed as follows: - All content that resides under any 'ee/' directory of this repository, if such directories exist, are licensed under the license defined in 'ee/LICENSE'. ... - Content outside of the above mentioned directories or restrictions above is available under the 'MIT Expat' license", followed by the full MIT text. The tier-setting license is MIT and the carve-out is named in the license itself. GitHub's license endpoint returns NOASSERTION for this repo, which is GitHub declining to classify the split file.

• https://raw.githubusercontent.com/Agenta-AI/agenta/main/ee/LICENSE recorded 2026-08-12

  The Agenta Enterprise License. The software "may only be used in production, if you (and any entity that you represent) have agreed to, and are in compliance with, the Agenta Subscription Terms of Service, available at https://agenta.ai/terms ... and otherwise have a valid Agenta Enterprise License." You may copy and modify it "for development and testing purposes, without requiring a subscription", and "it is forbidden to copy, merge, publish, distribute, sublicense, and/or sell the Software". Production use behind a paid license is functionality withheld from the MIT source.

• https://api.github.com/repos/Agenta-AI/agenta/contents/api/ee recorded 2026-08-12

  The carve-out is shipped code rather than a placeholder. api/ee holds LICENSE, __init__.py, databases, docker, src and tests, mirroring the MIT api/oss tree beside it (__init__.py, databases, docker, src, tests). web/ee and services/ee exist on the same pattern, web/ee carrying its own next.config.ts, package.json, public and src. The top-level ee/ directory contains only the LICENSE the root file points at.

agenta draws 15,013 PyPI downloads a month, inside the 10K-100K band, and has roughly 4,200 GitHub stars, which corroborate without setting the band. Early-adopter scale, the smallest in this category.

• https://pypistats.org/packages/agenta recorded 2026-08-13

  Downloads last month 15,013 for agenta (last week 3,250)

• https://github.com/Agenta-AI/agenta recorded 2026-08-13

  4,466 stars on Agenta-AI/agenta, the repository the agenta package is published from

The recorded feature list describes a product that has since been repositioned, and both readings are kept here. The older one is a broad feature matrix spanning prompt management, evaluations and OpenTelemetry observability, comparable to the Langfuse anchor at 4, with capability judged independently of Agenta's lower adoption. What the docs now open with is different: "Agenta is a workspace where you and your team build agents and automations. You do not need to know how to code." Against what this category grades, the whole remaining observability surface is tracing with request, token, latency and estimated-cost tracking. There is no evaluator catalogue anywhere in the docs, the prompt-engineering section 404s, and the OpenTelemetry, OpenLLMetry and OpenInference compatibility claims are gone; in their place is a multiplayer workspace with per-agent folders, 1000+ integrations and MCP servers, reusable skills, permissions with human approval, scheduled and event-triggered background agents, per-agent isolated sandboxes and version history. The rungs above 3 in this category require an evaluation layer, datasets and experiments, or prompt management on top of tracing, and Agenta 2.0 has none of the three; the rung below is for monitoring that does not trace individual model and tool calls, which this does. So 3, at medium confidence, with the honest caveat that the product is strong at something this category does not measure. No peer comparison is recorded, because nothing else in this category is an agent workspace and any target would assert a comparison class that does not exist. The open question is placement rather than band: a no-code agent workspace sits closer to agent orchestration than to telemetry, which is a recategorization for a human rather than a re-score.

• https://agenta.ai/docs/ recorded 2026-08-14

  "Agenta is a workspace where you and your team build agents and automations."; features listed are a shared multiplayer workspace, a shared folder per agent, "1000+ Integrations" via built-ins or MCP servers, reusable skills, human approval and permissions, background agents on a schedule or event, "Tracing and usage. Inspect every model and tool call.", and version history. No evaluator catalogue, no prompt-management section, no OTel/OpenLLMetry/OpenInference claim.

• https://github.com/Agenta-AI/agenta recorded 2026-08-14

  repo tagline and README now read "Agenta is a workspace where you and your team build agents and automations" and "The open-source workspace for building and running agents". The README no longer mentions prompt management, evaluators or LLM observability.

• https://agenta.ai/docs/ recorded 2026-08-14

  Beyond the features already recorded, the page carries harness choice across Claude Code, Codex and Pi, per-agent isolated sandboxes, and role-based access with SSO - and the whole of what this category grades is one line, "Tracing and usage. Inspect every model and tool call. Track requests, tokens, latency, and estimated cost."

• https://api.github.com/repos/Agenta-AI/agenta recorded 2026-08-14

  Repo metadata - description "The open-source workspace for building and running agents", corroborating the reposition from the repository side rather than from the marketing site.