FetchnGlue guide

Connect your applications with AI-assisted generation and human-controlled testing and release.

1. Set up your workspace

Sign in with your organization account, select your application and create an owned customer. Your administrator assigns author, reviewer and deployer access.

2. Describe and generate

Open AI Integrations. Select local CPU inference or explicitly choose a hosted AI provider. Supply API instructions, authentication method, objects, pagination and operations. Keep source credentials out of instructions. Inspect generated TypeScript, assumptions and validation results.

Local inference defaults to IBM Granite. A custom trained model must already be installed on your administrator-configured inference server and approved by its model ID. Your administrator reviews the model's origin, ancestry, permissive license and CPU requirements before enabling it. Entering a model ID does not install a model or choose a different server. Hosted providers remain explicit choices with their own credentials.

Use Work with the AI to request a refinement of the selected validated draft. Each turn uses the AI settings and API instructions you explicitly selected and creates a new draft. Inspect its TypeScript, rerun tests and obtain separate human approval. Conversation history stays in memory; saved drafts remain available after it is cleared. This view shows job status and completed results, rather than streaming model tokens.

3. Test and approve

Run contract checks, then validate the exact candidate against isolated source and destination accounts. Check counts, mappings, incremental behavior and retries. A separate reviewer approves the tested version; a deployer releases it through your controlled process. Standard local staging disables source release.

4. Connect and observe

Enter credentials using the owned connection flow for the released provider. Configure a destination and sync. Saving credentials does not prove access. Inspect Syncs, Logs and API Operations. Ambiguous writes require reconciliation.

Network access and support

Destination hosts must be reachable from your deployment. Ask your administrator for outbound IP addresses, TLS requirements and support. This deployment does not promise a hosted trial or a managed OAuth account.