Weldall CLI: Company systems for every terminal agent.
Your agent calls Jira, HubSpot or your own APIs with weldall request, using the signed-in user's permissions. It never sees a token.
$ npm install --global @weldall/cliThen run weldall login once. Requires Node.js 22.15 or newer. Prebuilt binaries for macOS, Windows and Linux are on GitHub.
you> Which Seabird contracts end this year?
$ weldall skills find "contract" $ weldall skills show contracts.manage
contracts.manage Manage contracts available
$ weldall request --scope contracts:read \ "https://contracts.example.com/api/contracts?customer=Seabird"
[ { "id": "contract-4711", "product": "Support Plus", "endsAt": "2026-12-31" }, { "id": "contract-4712", "product": "Licenses", "endsAt": "2027-06-30" } ]
pi> One contract ends this year: Support Plus on Dec 31, 2026.
Queried with your permissions (contracts:read).
Weldall logged the access.Works with your agent.
Any agent allowed to run commands, like GitHub Copilot or Codex. Plugins are available for:
- Claude Code
- Pi
- opencode
See Weldall live.
Watch an agent reach company systems through Weldall: signed in with the company’s identity provider and limited to granted scopes.
Neither the agent nor employees ever handle an API token or password.
No more API keys in .env files.
# .env
JIRA_API_TOKEN=atl••••••••7c2e
HUBSPOT_TOKEN=pat••••••••91ab
CONTRACTS_API_KEY=sk••••••••a71cOne key per service, on every machine, often with full access. Anyone who has it can use it.
$ weldall login $ weldall request --scope jira:read \ "https://jira.example.com/rest/api/3/search?jql=…"
Sign in once with your company SSO. Every request runs with your permissions. No keys in the repo.
How it works
- The CLI keeps the session and device key in the operating system’s secure credential store.
- The agent only calls
weldall requestand gets the response, never a token. - For each service, Weldall issues an ID-JAG, which is exchanged for a short-lived token bound to this device with DPoP.
What to know about Weldall CLI.
Which AI agents can use Weldall CLI?
Any agent that can run commands in its environment, such as Claude Code, Pi, opencode, GitHub Copilot and Codex. Claude Code, Pi and opencode have plugins, and the documentation shows how to install them. The CLI must be installed where the agent runs, and the agent must be allowed to run commands there.
What does the agent actually see?
Only the response data. The agent finds a skill, loads it and runs the calls the skill describes. For each call, the CLI checks that the service is registered and the scope is granted, sends the request itself and returns only the result. The agent never sees a token.
What is a skill?
Markdown instructions that tell the agent how to get a task done at your company, with the right weldall request calls, target URLs and scopes. Each person sees only the skills they’re authorized for. The agent finds them with weldall skills find and loads them with weldall skills show.
How do I connect my own API?
Secure your service with the TypeScript or Python SDK and publish a skill through it. An admin registers the service in Weldall as a resource with its own scopes and assigns them to people or groups. The skill then appears in the catalog for everyone authorized.
Can servers or CI jobs use Weldall without a person?
Yes. A machine gets its own client ID and key pair. With them, it gets a device-bound token that’s valid for five minutes, limited to the resources and scopes an admin has granted. The private key stays in your secret store, never in Weldall.
Do we need to expose all our APIs from day one?
No. Start with one clearly scoped API: register it as a resource with its own scopes and grant them to one group. IT expands access step by step as trust grows. Employees don’t see skills they aren’t authorized for.
For engineering teams