AI agents that do the actual work.
Not a chatbot stapled on. AuriusCRM's AI reads the BuilderJS schema, generates valid JSON, mutates automation trees through a typed tool layer, and lands every change inactive for human approval. Bring your own engine — OpenAI, Anthropic, Ollama (local), or our managed hub.
Every AI feature is a distinct task with its own prompt, engine policy, and rate limit.
No "general purpose AI". Each capability is a registered task with explicit inputs, outputs, rate limits, and an undo path. Audit-trail-grade by default.
Template generation
Generate full landing pages, email templates, funnels and forms from a prompt. Output: valid BuilderJS JSON, schema-checked.
page.generate · funnel.generate · form.generate · email.generate · block.rewrite
Text rewriting · in context
Inline rewrites for subjects, headlines, button copy, list names, segment descriptions. Each task keeps tone and never breaks layout.
subject · preheader · headline · cta · description · field-label · campaign-name · A/B subject · tag-suggest · spam-score
Support chatbox
In-app help, grounded in your KB. Multi-turn Q&A with KB citations, no tool-use — pure RAG. 20/min · 300/hour limits.
support.chat · backed by glossary + FAQ + screen context
Agent chat · with tool use
Multi-turn agentic chat. The agent has 60+ tools at its disposal, can mutate state, and reports every tool call.
agent.chat · 20/min · 500/hour · max 10 iterations per session
Inline rewrite · sparkle button
Highlight any text in the builder, click ✨. Choose: shorter · bolder · more formal · custom prompt. Replaces selection in place.
shorter · longer · bolder · formal · informal · custom
Every task is logged
Each request stores: input · output · engine · model · tokens · latency · cost · user · timestamp · feedback (👍/👎). Daily rollups in the admin dashboard.
ai_requests · ai_tool_calls · ai_messages · ai_feedback · ai_daily_rollups
Your AI bill. Your data residency. Your model choice.
Run cheap tasks on Ollama (local, air-gapped, free). Run quality-critical tasks on Claude or GPT-4. Swap engines per-task without touching prompts. Bring your own API key, your own on-prem deployment, or use the managed AcelleAI Hub.
Per-engine API-key verification, per-engine rate limits, per-task model overrides. Switch engines mid-flight; in-flight tasks finish on the old engine, new tasks pick up the new one.
The agent doesn't have free hands. It has a typed toolbox.
Every action the agent can take is a registered tool with a JSON schema. The schemas are embedded in the system prompt; tool calls are validated before execution; mutations land INACTIVE for review.
AI never deploys directly to production. You do.
Every AI mutation — a generated page, an automation tree, a new segment — lands INACTIVE. The agent finishes with a report: "here's what I built, here's where to review, here's the undo button". You activate. Or you don't.
Every request, tool call, latency reading, token count, and 👍/👎 is persisted in five database tables. The admin dashboard shows daily rollups. Compliance teams can audit every prompt and every action the AI took, who approved, who reverted.
Requests · last 24h
847
↑ 12% vs last week
Avg latency
1.2s
P95: 3.4s · p99: 6.1s
Tokens used
614K
in 412K · out 202K
Total cost
$3.42
openai · 89% of spend
Approval rate
94%
of AI mutations activated
Undo rate
3.1%
of activated changes reverted
Generate. Review. Approve. Repeat.
The Copilot is built into every AuriusCRM workspace from day one. Open the playground to see the inline rewrite (sparkle button) in action.