Swytchcode Examples: six demos you scaffold, eleven you clone

The first six scaffold, the rest are meant to be cloned

A demo repository can be read two ways, and this one splits its contents along that line. The first six examples are reachable through the `swytchcode examples` generator, which scaffolds them fresh with dependency management already in place. Everything after that is described as a reference architecture meant to be explored and cloned directly, with no CLI path in front of it.

The difference in approach matters more than the count. A scaffolded project hands you a dependency graph and environment configuration you cannot inspect until it is on disk. A cloned reference architecture shows every file, including the framework glue and the assumptions it makes. Teams evaluating an agent workflow tend to want the second, and teams shipping something this week want the first.

Where the line falls is stated rather than inferred. `openclaw-swytchcode` is the sixth entry, and it is also the first one written in Go rather than TypeScript.

Install the CLI, generate, then the example's own README takes over

Three commands stand between a fresh machine and a scaffolded demo. Node.js is the only prerequisite named:

npm install -g swytchcode

The generator is interactive, and you pick on two axes: a use case such as Customer Onboarding or Bug Escalation, and a framework such as LangGraph.

swytchcode examples

From there you are handed back to the generated directory and to that project's own documentation:

cd <your-scaffolded-demo>

That last step is where the documentation thins out. The top-level README reproduces no setup steps, environment variables or credentials for any example, so the generated project is the only place they can be. Comparing six examples means reading six different READMEs written by the same authors in the same repository.

The directory table names thirteen examples and the tree holds seventeen

The tree carries seventeen demo directories and the table indexes thirteen. Four are not in the table at all: `gmail-assistant-anthropic-typescript`, `gmail-google-calendar-openaisdk-demo`, `langgraph-google-calendar` and `zoom-swytchcode-demo`. Three of those four are mail or calendar integrations, which reads more like an unfinished index than a decision to keep Google's APIs out of the list.

Link text and directory names do not always line up either. The row labelled `stripe-subscriptions` points at `./Stripe-Subscriptions`, and every LangGraph example carries a `-langgraph` suffix in the tree that the table hides behind a short label. None of that breaks a clone by hand, but a script that walks the table to find directories will miss four projects and mismatch one name.

Tags name HubSpot and Stripe, and no credential is documented anywhere

Every row in the table ends in tags, and those tags are the only place the integrations are named: HubSpot, Stripe, Resend, Jira, Slack, Sheets, Notion, GitHub. That is enough to tell you the shape of a demo and not enough to run one. No section of the README names an environment variable, an OAuth flow, a webhook URL or a sandbox key, and the CLI is described as providing templates pre-wired with environment configurations without showing what those configurations contain.

For a payments or CRM workflow that gap decides whether the example is a starting point or a sketch. `create-and-send-payment` is billed as generating and sending payment links, `stripe-subscriptions` covers the subscription lifecycle including dunning, and `weekly-reporting` distributes analytics through Sheets and Notion. The parent project claims built-in idempotency, audit trails and strict policy enforcement as platform features. Nothing in this repository shows the wiring that would let you confirm any of it.

Five agent frameworks in a repository whose language is TypeScript

Five frameworks appear across the demos, and a TypeScript repository heading hides two of them. LangGraph covers the largest group, including `langswytch`, listed as advanced LangGraph cognitive patterns with multi-agent, orchestration and memory tags. Two examples use the Swytchcode SDK directly, `github-issue-integration` for automated GitHub issue creation and `stripe-subscriptions` for billing.

Then the languages change. `openclaw-swytchcode` is a GitHub issue triage bot built on OpenClaw in Go, tagged GitHub, automation and reliability, and `swytchcode-refund-agent-openclaw` is a second OpenClaw project for refunds. The last two are Python: `star-repo-openai-agents-python` stars a repository through the Swytchcode CLI on the OpenAI Agents SDK, and `github-agent-crewai-python` pairs CrewAI with GitHub operations. One `npm install -g` therefore yields a generator that will hand you projects with different runtimes and dependency managers.

Money-touching demos sit next to the easiest ones to adapt

The examples closest to real consequences are also the ones with the most setup ahead of them. `create-and-send-payment` generates and sends payment links through Stripe and Resend, `stripe-subscriptions` handles the subscription lifecycle with dunning, and `swytchcode-refund-agent-openclaw` automates refund processing. Beside them `fintech-compliance-multiuser` covers multi-user compliance workflows, the one row in the table whose subject is a control rather than an integration.

Customer and sales work reads as the opposite. `customer-onboarding` covers B2B onboarding and billing setup, `lead-qualification` does lead enrichment and CRM routing, and `bug-escalation` triages across GitHub, Jira and Slack. Those are the closest thing here to templates a team could adapt in an afternoon, and they are also the ones whose credentials are least visible from the repository root. Engineering triage through three systems is a reasonable first clone for exactly that reason: it touches integrations most teams already run.

MIT licensed, one release tag, and a .DS_Store in the root

Swytchcode presents itself as the layer between autonomous agents and real APIs, databases and tooling, with native SDKs for LangGraph, the OpenAI Agents SDK and the Vercel AI SDK, and a homepage at swytchcode.com. This repository carries the MIT license and a CONTRIBUTING.md. Its one release is tagged `latest` and dated 2026-07-03, and the last push to main was on 2026-09-10.

A `.DS_Store` file sits in the repository root, which is the sort of detail you only see in a repository maintained by hand rather than by a release pipeline. What a reviewer will not find in any of these projects is a runnable demo with no keys attached. Every example needs an account at a third-party service, and the documentation for those accounts lives with those services.

Editorial conclusion

Adopt this repository if you are choosing an agent framework and want existing wiring for HubSpot, Stripe, Jira or Sheets in one place, since the framework spread and the integration tags are the useful part. Do not adopt it as a reference for credentials or deployment: no example's environment variables, keys or run steps are documented at this level, and four directories are missing from the index table. Start by cloning the reference architecture whose tags match a service you already pay for, then read that project's own README, because it is the only setup documentation in the project.

Frequently asked questions

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. swytchcodehq/swytchcode-examples on GitHub