AI Employees and n8n workflow automation are now open

[antonyrag]

AI Employees and n8n workflow automation are now open

Two pieces that were previously hosted-only are now part of ragleap-core, single-tenant, MIT-licensed, same as everything else in this repo.

AI Employees (#130, #133)

Nine role-based agents (manager, secretary, CEO, sales, support, HR, finance, marketing, operations) — each with its own personality, and each with real, persistent, pgvector-backed memory that learns from owner instructions, conversations, and uploaded documents over time. Not a static prompt — the business profile and learned context genuinely rebuild as the deployment gets used.

Wired into /chat and /chat/stream via an optional role param:

POST /chat?question=...&role=support

core/employees/ for the module, core/employees/skills.py for the retrieval logic if you want to see how role context gets built.

n8n workflow automation (#141)

Configure a webhook, pick which channels should trigger it, and it fires automatically after the AI replies — no code required per automation. Works today on WhatsApp, Telegram, and Discord.

POST /n8n-workflows
{"name": "...", "webhook_url": "...", "channels": ["whatsapp", "telegram"], "is_active": true}

core/workflows.py for the module — it's a genuinely small, clean single-tenant port of the same trigger logic running in production.

What's still hosted-only, and why

Not everything is open, and the reasoning is consistent: things that are genuinely SaaS infrastructure (multi-tenant billing, team permissions, audit logging, managed hosting) stay commercial. Things that are actual engineering — like the two above — are real candidates to open, one at a time, verified working before merging, not just wired-and-assumed.

Manager AI is the next big candidate under consideration, but it's explicitly not scoped yet — it's entangled with genuinely SaaS-only pieces (analytics, team permissions, database connections) and needs its own decision before any code starts. Persistent cross-channel memory is a similar story — already audited once and found architecturally fused with billing in a way that isn't a clean extraction.

Want to contribute?

Both features shipped with immediate follow-up work already scoped and labeled:

• #134 — test coverage for /chat with-role and without-role cases (in progress, thanks @aayushi-tomar)
• #135 — example scripts for AI Employees
• #143 — wire n8n triggers into the Voice channel

All labeled good first issue, all with a clear reference implementation already in the codebase to work from.