Chatbots, controlled

Right model. Right Skills.
Right audience.

Deploy distinct chatbots to distinct teams, all from one admin panel. Engineering gets Claude with GitHub and Linear Skills. Customer support gets a Zendesk-aware bot on a low-cost model. Executives get a citation-grounded assistant pinned to leadership docs. One product, many configurations, one billing relationship.

Start Free Trial See how it works

14 days free — no credit card required

The problem with "enterprise AI"

One chatbot for the whole company doesn’t work.

Microsoft Copilot gives everyone the same Copilot. ChatGPT Enterprise gives everyone the same ChatGPT. But engineering needs a code-aware bot with GitHub access. Support needs a Zendesk bot that knows your help-center voice. Execs need citations and source links, not vibes. And finance can’t use a chatbot that hasn’t been told what it’s allowed to discuss.

Nexil treats chatbots as a config surface, not a product. Every chatbot is an agent. Every agent has its own model, its own Skills, its own data sources, its own response governance, and its own access list. Roll out as many as you need.

✕ Everyone gets the same chatbot

✕ No per-team model choice

✕ No per-team tool access

✕ No per-team governance

✓ Nexil: each team gets its own chatbot

The rollout pattern

Three knobs per chatbot

Every chatbot in Nexil is built from the same three configuration choices — mix and match to fit each audience.

1. Pick the model

Choose from the production LLMs hosted in your tenant — GPT-4o, GPT-4.1, Claude (Opus / Sonnet / Haiku), Llama, Grok, Mistral, Cohere — or bring your own (Azure OpenAI, OpenAI direct, Anthropic). Per-1M cost is shown inline so admins can balance capability against spend.

2. Snap on Skills

Attach the capability packs the audience needs. A support bot gets the Zendesk Skill. An engineering bot gets GitHub + Linear + Atlassian. A finance bot gets the SQL Analytics Skill. Drop in any SKILL.md from the wider ecosystem or write your own internal ones.

3. Scope the audience

Map an IDP group (Entra security group or Clerk role) to the agent’s access list. Only those people see it in the portal. Per-agent ACLs sit on top of tenant-level access — admins control who can use what, no Nexil-specific user database to maintain.

Example rollouts

What a real rollout looks like

A mid-market company isn’t buying one chatbot — they’re standing up four or five, each tuned to its team. Here’s a representative deployment.

Engineering bot

Model: Claude Sonnet 4.6
Skills: GitHub, Linear, Atlassian, SQL Analytics
Audience: Engineering AD group
Style: Concise, technical, code blocks expected
Sources: Confluence engineering space + repo READMEs

Customer support bot

Model: GPT-4o mini (cost-optimised)
Skills: Zendesk, Web Research
Audience: Support team + external customer portal
Style: Friendly, empathetic, on-brand voice
Sources: Help center articles + product docs

Executive assistant

Model: Claude Opus 4.7
Skills: Microsoft 365 (mail + calendar), SharePoint
Audience: Leadership Entra group only
Style: Citation-grounded, sourced answers
Sources: Board decks, strategy docs, exec SharePoint

HR policy bot

Model: GPT-4.1
Skills: none (chat only)
Audience: All employees
Style: Plain language, must cite the policy doc
Sources: Employee handbook, benefits PDFs

Finance bot

Model: Claude Sonnet 4.6
Skills: SQL Analytics, Stripe
Audience: Finance + Accounting groups only
Style: Numbers-first, no speculation
Sources: Tenant data warehouse (read-only)

Sales coach

Model: Claude Sonnet 4.6
Skills: Salesforce, HubSpot, Slack
Audience: Sales org
Style: Coaching tone, asks clarifying questions
Sources: Sales playbooks + win/loss notes

Same platform. Same admin panel. Same billing. Six different chatbots, each tuned to who’s using it.

Per-agent governance

Tell each chatbot how to behave

Response governance is per-agent, not per-tenant. The finance bot can be told never to speculate. The customer-facing bot can be told to always end with a help-center link. The engineering bot can be told to default to code blocks. No prompt copy-paste between agents.

Tone, length, language, persona — set per agent, applied to every response
Prohibited topics — the chatbot refuses with a custom fallback message
Required disclaimers — legal, compliance, or policy lines auto-appended where needed
Grounding mode — strict (only from sources), prefer-sources, or open (general knowledge allowed)
Few-shot examples — show the agent what a good answer looks like for your domain
Starter prompts — guide first-time users to the right questions instead of a blank box
Live preview pane — test the draft agent right next to its config before saving

The Nexil agent configuration editor with live preview

The agent management table inside a Nexil tenant

Model freedom

Use ours, or bring your own

Your tenant ships with production LLMs hosted for you — no Azure subscription, no API keys, no infrastructure. Or, on Business and Dedicated tiers, point an agent at your own Azure OpenAI deployment, OpenAI direct, or Anthropic account. Per-1M cost is shown inline next to every model in the picker.

GPT-4o, GPT-4.1, GPT-4o mini
Claude Opus 4.7, Sonnet 4.6, Haiku 4.5
Llama 3.3 70B, Llama 4 Maverick
Grok 4 Fast, Grok 3
Mistral Large 3, Cohere Command-A
Bring your own — Azure OpenAI, OpenAI direct, or Anthropic (Business toggle, Dedicated default)

Self-service admin

Non-technical admins build agents

No prompt engineers, no platform team, no developer involvement to roll out the seventh chatbot. Department heads and ops leads create and manage agents from the admin panel.

Visual agent editor

Edit prompt, model, Skills, sources, and ACLs in a form — not in YAML. Live preview shows how the agent responds before you save.

Delegated admins

Per-agent delegated admins from your IDP. The finance team owns the finance bot. The support team owns the support bot. IT doesn’t have to be in the loop.

Versioning

Every edit snapshots the prior version. Roll back if a prompt change makes the bot worse. Audit who changed what and when.

Lock-out guard

You can never accidentally remove your own admin access. Tenant admins manage everything without ever calling support.

Right model. Right Skills.Right audience.