AI agent

Every autopilot escalation auto-generates a post-mortem with root cause, missing-article suggestion, and a guardrail change you can apply in one click.

Reviewers grade past AI replies on accuracy, voice, and helpfulness. Their notes feed the learned library and tune the AI.

The AI gets smarter from your past conversations. High-confidence Q+A pairs get promoted into a library it can lean on.

Every AI reply (sent or drafted) is logged with model, tokens, cost, confidence, sources, and action taken.

Try the AI on real or made-up messages without sending anything. See drafts, sources, confidence, and where coverage is thin.

Notion and GitBook indexed into embeddings, queried via RAG, visible as a similarity graph between articles.

Set a monthly AI budget. Get alerts at 50, 75, and 100%. Auto-pause at 100% so you don't get surprised.

The set of rules that keeps the AI out of conversations where it shouldn't speak.

Confidence is the dial that turns AI from cautious to aggressive. Two floors and a fallback chain.

Every new conversation is stamped with topic, priority, and a confidence score. Filters and routing rules use these.

Cap how long the AI's drafts get and how it handles long threads. Shorter is almost always better.

Teach the AI how your team writes. Sample replies beat adjectives, every time.

Three reply modes shape how visible the AI is in your inbox. Pick one per channel, or layer them.

One toggle that promotes the AI from teammate to first responder. New tickets get answered before a human looks.

Connect both Anthropic and OpenAI keys so a provider outage doesn't stop your inbox.

Sonnet 4.6, Haiku 4.5, or GPT-5.4: when each one wins, what they cost per resolution, and how to mix them.

Connect your Anthropic and OpenAI API keys so Ochre's AI runs on your account, at provider cost, with no Ochre markup.

How Ochre's AI agent reads, drafts, labels, and routes tickets, plus where the human stays in the loop.