The grounded AI assistant that answers customers wherever they ask. Run it on your inbox, your website, or both — same KB, same tone, same audit trail. Strict mode by default: the bot will not answer anything not in your docs.
Support inboxes are a queue of repeats. The answer is already in your docs — a policy, a product spec, a how-to. But the human in the loop still has to find it, phrase it, and paste it. Meanwhile first-response time slips, tone drifts, and nobody can audit why a reply said what it said.
Same knowledge base, same tone, same audit trail. Orepli meets customers where they are — in your inbox and on your site — and routes the rest to a human with a clean summary of the conversation so far.
Connect Gmail or IMAP. Every inbound message is classified, grounded against your KB, drafted, and either auto-sent above your confidence threshold or queued for review. Cited, logged, and reversible.
Drop a one-line script on any page. Visitors get a grounded assistant that answers from the same KB your mailboxes use. When it can't help, the visitor escalates — Orepli emails your team a clean summary of the chat.
<script src="https://orepli.com/widget.js" data-orepli-key="pk_live_…" async></script>
Every email runs through four gates: connect, ground, classify, reply. You keep control at each one. Turn auto-send on when you're comfortable. Leave it off forever if you prefer every reply reviewed.
Run Orepli on email only, on a chat widget only, or both — they share one knowledge base. Connect Gmail in a click, plug an IMAP mailbox, or generate a one-line embed snippet for your site.
Drop in FAQs, policies, SOPs, product docs. Orepli chunks, embeds, and scopes them to whichever channel needs them. Copy a mailbox's KB into a widget for an independent snapshot.
Strict KB mode is on by default — the chat refuses anything not in your docs. For email, set a confidence threshold for auto-send. Pick a tone per mailbox. Start in draft-only until you trust it.
Every reply cites its KB chunks. Every Claude call shows up in your usage page. Skipped emails (newsletters, bounces) get a reason. Hard cap at 2× plan tokens — admin-tunable per tenant.
Built on the stack that serious RAG teams already run — Postgres with pgvector, Claude for generation, explicit citations, per-tenant isolation. No black box.
Every draft cites the KB chunks it was built from. pgvector under the hood. Swap the embedder if you need HF, OpenAI, or local.
Replies only auto-send when model confidence clears your threshold. Everything else becomes a draft or a routed escalation.
Per-mailbox voice — terse, warm, formal, technical. Different queues, different voices, same knowledge base.
Edge cases forward to the right human with the reason, the retrieved context, and the draft we would have sent.
Received → classified → retrieved → drafted → sent. Every step is logged, queryable, and isolated per tenant.
Failures never silently disappear. Retry, inspect, or escalate. Your ops team gets a real picture of what broke and why.
One-line embed on any site. Shares the mailbox's KB and tone. Per-widget origin allowlist, rotatable public key, AI-summarised escalation to a human.
Email traffic and chat traffic both bill against the same monthly token pool. One invoice, one quota, no surprise line items.
Cited chunks, confidence score, model, latency, token count. When a draft ships, you can trace back why — and when it doesn't, you see exactly which rule caught it.
Pick the plan that fits your mailbox count and monthly token budget. Overage is linear — no surprise bills, no throttles. Upgrade or downgrade any time.
For a founder or a small team automating a single support queue.
For small support teams running separate queues for sales, ops, and support.
For teams scaling RAG-native email across brands or regions.
Still have questions? sales@orepli.com — we reply with Orepli, obviously.
Only when the draft's confidence score is above the threshold you set on the mailbox. Everything under the threshold becomes a reviewable draft or gets escalated. You can also run a mailbox in draft-only mode indefinitely.
Gmail via OAuth, or any IMAP/SMTP inbox with TLS. Credentials are encrypted at rest with a tenant-scoped key. OAuth tokens are refreshed automatically.
Claude from Anthropic by default, with a mock mode for local development. Embeddings default to a deterministic local hash embedder; you can switch to Hugging Face sentence-transformers or a managed provider via a single env var.
Knowledge base documents are scoped to the mailbox you upload them to. Email content is stored per tenant in Postgres with pgvector. You can request export or deletion at any time. See the Privacy page for the full picture.
Yes. Register a workspace, connect a mailbox, run inbound emails through the simulator, and preview drafts before turning on auto-send. No card needed to evaluate.
Yes, for Scale customers and above. The stack is FastAPI + Postgres + pgvector + Next.js, deployable via Docker Compose. Contact sales@orepli.com to discuss.
Create a widget in admin, attach it to a mailbox, and you get a one-line embed script with a public key (pk_live_…). Drop it on any page. The widget answers from that mailbox's knowledge base and tone. Per-widget origin allowlist blocks abuse, the key is rotatable, and visitors can escalate — Orepli emails your team a clean AI-generated summary of the chat.
Yes. Email traffic and chat traffic share the same monthly token pool on your plan. No separate SKU, no per-conversation surcharge. Widget calls are tracked against your usage the same way email calls are.
Spin up a workspace, connect a mailbox, upload a KB, and watch the pipeline work on your real inbox. No credit card to evaluate.