neutron AIConnect the workflow you already use
A support preference should reach the next authorised task without making you replace your model provider. Neutron AI exposes HTTP calls for this integration. There is no native no-code connector promised here. Use developer help if your automation cannot keep secrets or enforce per-user access.
Trusted HTTP automation recipe
- Follow the quickstart and store the Nucleus token in the automation service's encrypted credential field. Never place it in a browser form, a webhook URL or workflow output.
- Authenticate the incoming user in your backend. Map that identity to a fixed permitted Nucleus and Scope. Do not copy
scopeIdornucleusIddirectly from an untrusted webhook body. - Add an HTTP POST step to
https://neutronai.dev/api/v1/memory/remember, with bearer credentials, JSON content type and the synthetic quickstart body. Expectaccepted: trueand savememoryId. - Before the existing model step, POST to
/v1/memory/agent-contextwith the authorised Scope, task and token budget. PasspromptContextas supporting context to that model step. - Add a separate correction/consent-revocation path that calls
/v1/memory/forgetwith the exact Nucleus, Scope and memory ID. Remove any copied prompt context in your own workflow cache. - Stop on 401/403/402. Use bounded retries for transient failures; avoid automatically replaying billable writes after an ambiguous timeout. Keep tokens and private memory out of execution logs.
Check whether your automation provider supports protected credentials, HTTP steps, conditional error branches, deletion events and private log retention before choosing it. This recipe does not certify any third-party connector, plan or configuration.
Framework boundaries
| Stack | Where to make the call | Where to use the result | Responsibility retained by your app |
|---|---|---|---|
| Node.js / Vercel | Server route or server action using secret environment variables | Before the existing model request | User session, Scope mapping, timeout and 402 handling |
| Python / LangChain | A server-side callable before the model or chain invocation | Add promptContext to the chain's supporting context | Do not treat retrieved text as new system instructions |
| Python / LangGraph | A node before model execution, followed by a bounded state update | Supply the pack for that task | Keep checkpoint state, current task and Neutron memory boundaries distinct |
| PHP / Laravel | Service class called after authentication/authorisation middleware | Add context before your model gateway call | Never put the bearer token in a Blade page or public job payload |
| Go | A server-side client with request context and timeout | Merge eligible context before the model HTTP request | Check errors and retain cancellation |
| Rust | A server-side async client or service adapter | Add eligible context to the chosen model request | Keep secrets, access policy and execution decisions in the service |
The complete language examples use public REST directly, so they work without waiting for a framework plugin or typed SDK parity. They are backend examples, not automatic integrations with every framework version. Measure setup effort on your application rather than relying on a universal time-to-value promise.