Skip to content
Help guide the future of agentgateway — Take the survey

For the complete documentation index, see llms.txt. Markdown versions of all docs pages are available by appending .md to any docs URL.

Skill for AI assistants

Page as Markdown

Install a skill that teaches your AI coding assistant, such as Claude Code or Codex, to set up, configure, and verify standalone agentgateway for you.

Let your AI coding assistant set up agentgateway for you. Unlike the other getting-started guides, this guide does not route traffic through agentgateway to an AI backend. Instead, you install a skill: a packaged set of instructions that teaches your assistant, such as Claude Code or Codex, how to install, configure, and troubleshoot agentgateway for your edition and deployment mode.

After you install the skill, your assistant reads the current public documentation, inspects your actual runtime, and verifies the behavior that you request.

Target: OSS standalone configuration. Run agentgateway on your selected reachable host, VM, server, or container environment. A laptop is one possible target. Standalone Helm hosting also uses this configuration model. For Gateway API and controller-managed resources, use the Kubernetes skill guide.

Note

These skills are a preview. Review generated changes and verify them in your environment. To set up agentgateway yourself instead, follow the other getting-started guides.

Why use a skill?

The skill gives your assistant a repeatable workflow: identify the deployment, retrieve applicable documentation, make the requested change, and check the result. It separates configuration generation from a running deployment and keeps OSS and Enterprise capabilities explicit.

It has no fixed gateway release requirement or expiry date. Your assistant discovers the actual runtime and selects the corresponding documentation and schemas. A new gateway release does not require a new skill release. Public documentation and workshops provide the source material; no Solo account or internal support tools are needed. Public Solo documentation search is optional.

Installing a skill adds instructions to your assistant. Gateway changes happen when you request them, through the assistant’s existing tools and permissions.

Before you begin

Use Claude Code, Codex, or another assistant that can load a complete SKILL.md folder. Setup needs file/shell access to the selected environment and access to public HTTPS documentation. Choose your provider and credential source; keep secret values out of prompts.

Identify the target host or container environment, how your assistant can access it, the existing configuration owner, and the network from which clients and your browser connect. API and UI addresses must be reachable from that network. An administrative UI may use a protected hostname or an authorized tunnel; it does not need to be exposed publicly.

Install and load the skill

From your project directory, use a current Node.js LTS to install the selected skill from agentgateway.dev. Review the installer’s proposed changes. The commands follow the website’s current skill content; there is no skill or gateway version to bump. You can also install from the source repository when working on the skill itself.

npx skills add https://agentgateway.dev --skill agentgateway-oss-standalone --agent claude-code

These commands install into the current project. Add -g only for a personal installation across projects. Codex uses .agents/skills, which other compatible assistants can also discover.

ZIP alternative

Download the current agentgateway-oss-standalone package. Checksums and content fingerprints identify the exact contents without dates or release numbers. Extract the whole folder, including references/, scripts/, and package.json.

mkdir -p .claude/skills
unzip -n "$HOME/Downloads/agentgateway-oss-standalone.zip" -d .claude/skills

For an existing installation, use the update instructions below instead of merging files into an older folder.

Load and check the skill

Open your assistant in the project and enter the following at its prompt:

/agentgateway-oss-standalone Identify your edition, deployment mode, content fingerprint, and documentation sources. Do not change the deployment yet.

If discovery fails, check that the skill directory directly contains SKILL.md and start a new session in that project.

Set up your first provider

Choose the provider you use. The assistant should resolve missing target/provider inputs, preserve existing configuration, run one bounded request, and return the verified API endpoint and protected UI access path. A test request can incur normal provider charges.

OpenAI

Set up agentgateway in my selected deployment environment for OpenAI.
Use my chosen model and existing provider secret source; ask for either if missing.
Preserve the configuration owner. Verify a request through the gateway and show
me the reachable API endpoint, protected UI access path, and rollback commands.

Gemini through Vertex AI

Connect Gemini through Google Cloud Vertex AI in my selected deployment environment.
Confirm the Google Cloud project, region/location, model, and workload identity or ADC.
Use Vertex AI authentication, rather than a Gemini Developer API key.
Verify a request from the actual gateway workload and show API and UI access details.

Amazon Bedrock

Connect my Bedrock models in the selected deployment environment.
Confirm AWS region, model or inference-profile IDs, and the gateway's IAM identity.
Verify a request through the gateway and return the reachable API endpoint,
protected UI access path, and stop or rollback commands for these changes.

For preparation only, begin with “Prepare the files without deploying or starting anything.” If access or credentials are missing, the assistant should finish independent preparation and identify the checks it could not perform.

Check versions and behavior

Ask the assistant to inspect the actual gateway version before validating configuration. In controller mode it also checks the deployed controller, proxy images, and CRDs. The skill has no maintained release allowlist. It uses the discovered runtime to choose documentation and validation methods, preserving your existing deployment version.

The optional bundled helpers need Python 3.10+. They inspect package integrity/runtime evidence, classify captured MCP tool responses, and check controlled masking examples. Results distinguish a passed check, failure, and an inconclusive result. Neither a version observation nor valid YAML proves a working deployment.

Continue with your own tasks

Independent identities and budgets

Give five agents separate identities and a $10 budget each per fixed 24-hour window.
Check what my edition and deployment mode can enforce. Verify independent accounting,
model prices, persistence, and a bounded enforcement test. Keep credentials out of chat.
If the requested dollar enforcement is unavailable, explain the gap before choosing an alternative.

Standalone and controller cost-control surfaces can differ. A token quota is not an exact dollar allowance, and post-response charging/concurrent requests can overshoot a threshold. Consult the current budget documentation for your deployment before relying on enforcement.

JWT identity and MCP code mode

I want budgets tied to authenticated JWT identities and MCP code mode.
Check native OSS support for each requirement in this deployment. Distinguish
claim-keyed token limits from dollar enforcement. Keep implementation OSS and
explain any relevant Enterprise capability with a public source.

A relevant Enterprise availability note is informational. It does not change editions or install licensed components. The skill checks capabilities separately rather than treating all JWT or MCP features as Enterprise-only.

Troubleshooting

Requests through my gateway fail. Inspect the actual runtime and configuration,
identify the failed layer, make the smallest authorized correction, and verify it.
Preserve unrelated settings and report any remaining unrun checks.

Update, remove, or share

For repository installations, inspect installed packages and update the selected skill:

npx skills list --json
npx skills update agentgateway-oss-standalone --project

Back up local customizations outside skill directories first: updates can replace them. Start a new assistant session afterward. Use --global instead of --project for an intentional personal installation.

For manual ZIP installations, download the current ZIP and replace the complete folder after backing it up. Keep its content fingerprint with any bug report. Sharing that ZIP gives another person the same reviewed instructions without requiring a repository commit.

npx skills remove agentgateway-oss-standalone

This removes the selected skill across project clients. Client-specific removal can leave .agents/skills present because several clients share it. Removing a skill does not stop a gateway or remove its configuration, credentials, resources, or accounting data.

For feedback, open a website issue with the content fingerprint, assistant/runtime information, deployment mode, sanitized request, and observed result. Keep credentials and customer data out of the report.

Was this page helpful?
Agentgateway assistant

Ask me anything about agentgateway configuration, features, or usage.

Note: AI-generated content might contain errors; please verify and test all returned information.

Tip: one topic per conversation gives the best results. Use the + button in the chat header to start a new conversation.

Switching topics? Starting a new conversation improves accuracy.
↑↓ navigate ↵ select esc dismiss

What could be improved?

Your feedback helps us improve assistant answers and identify docs gaps we should fix.

Need more help? Join us on Discord: https://discord.gg/y9efgEmppm

Want to use your own agent? Add the Solo MCP server to query our docs directly. Get started here: https://search.solo.io/.