Algen Release Guard
Explains deterministic GitHub readiness findings with evidence IDs
Loading
Loading
Loading
Loading
Loading
Loading
Loading
Loading
Loading
Loading
Loading
Loading
Discover, evaluate and adopt AI agents as versioned releases, with visible evidence, declared permissions and clear delivery paths.
Browse all agentsInspect purpose, boundaries, and release details before you choose.
Explains deterministic GitHub readiness findings with evidence IDs
Algen's reference support agent for a small online shop. Answers questions from help articles, looks up orders and asks a person to approve every refund. Built by Algen.
Start with the task, then narrow by category and framework.
Review assessment scope, permissions, limitations, and setup needs.
Follow the declared delivery path for the exact version you reviewed.
How to evaluate, compare and publish AI agents, in plain language.
A practical AI agent maturity model and matrix: 40 capabilities in 12 areas, rated 0 to 3 with the strength of their evidence, mapped to stripes from 0 to 7.
Read the guide 02An AI agent safety checklist and matrix: secrets, prompt injection, code execution, approvals, egress, run limits and error handling, with what good looks like for each.
Read the guide 03The difference between an AI agent demo and a production-ready agent: written scope, typed tools, bounded runs, honest errors, approvals, evals and tracing.
Read the guideAlgen Agent Hub is a governed AI agent marketplace. Every listing is a versioned agent release with declared permissions, a clear delivery path and its AgentStripes result, so teams can evaluate an agent before they adopt it.
A directory links to tools. Agent Hub lists immutable, digest-pinned versions and attaches evidence to the exact version you are reviewing. Each version carries its AgentStripes result with the code it cites, ratings come only from users with a recorded acquisition, and each listing states how it is delivered.
AgentStripes is Algen's maturity model for AI agents. Three model judges rate 40 capabilities from an agent's code against written definitions, and every rating cites the code or an instrument finding. Each rating, weighted by severity and by the strength of its evidence, adds to a score, and every 12.5 points earns a stripe, up to 7. Three gates can hold an agent lower. Stripes 1 and 2 can come from the code; 3 to 7 need the safety behaviours shown in runs. It is not certification or a safety guarantee.
No. Stripes describe how one version of an agent is built and what evidence backs it. They are not certification or a guarantee. Review permissions, data use and the cited evidence for your own context before adopting any agent.
Agents for research, customer support and other business and engineering work, built with frameworks such as Algen Runtime and LangGraph. Agent Hub is framework- and cloud-neutral, so CrewAI, OpenAI Agents SDK, custom containers and hosted endpoints share the same review boundary.
Each listing declares one delivery path: an Algen package you download, verify against its SHA-256 digest and run locally; a source container; or a publisher-hosted agent. Agent Hub does not execute the package or grant access to your files.
Yes. Publishing and acquiring agents are free in the first phase; paid transactions are deliberately deferred.
Only users with a recorded acquisition. Each rating stays tied to the version that person acquired, and each user has one editable rating per agent.
Recommends support category, urgency, risk flags, and queue while requiring a human decision.
Produces source-linked local document findings and requires a human decision before release.
Explore information with findings you can trace back to their source.
Sign in with a passwordless email link, declare ownership, license, data use, network access and delivery, then upload an immutable package. Check your agent with AgentStripes first: a version needs at least two stripes. The package is quarantined and inspected, an administrator reviews it, and it is published as a version.