No owner, no scope
AI launches as a company-wide ambition instead of a specific job to be done. The pilot demos well, changes nothing, and quietly ends.
With AgenticliiEach agent has one job, one business owner and a baseline to beat.
AI agents for Microsoft 365
Add the expertise.
Not the headcount.
Agenticlii agents take on specific jobs, from executive assistant to HR, finance and project management. They work inside the Microsoft 365 your people already use, behind seven layers of security, and run on whichever AI model you prefer, with very little asked of your IT team.
Works in
OutlookTeamsSharePointOneDrivePlannerExcel
HR
HR & People AgentSample chat · Teams
MK
A new hire starts on the 18th, which is mid benefit cycle. What do I need to do?
HR
Here is our process for a mid-cycle start:
- Submit the new-hire form by the 11th so payroll can prorate deductions.
- Benefits start on the 1st of the month after 30 days of service.
- Send the enrolment pack on day one. The employee has 31 days to elect.
Employee Handbook · §4.2Benefits Guide 2026 · p.7
Signed in with your directoryAnswer from approved sources onlyLogged for audit
The problem
Everyone is piloting AI. Far fewer are deploying it.
Most AI initiatives stall for one of three reasons, and none of them is about what the model can do.
Copy, paste, and hope
To get help from AI, staff copy employee, customer and financial details out of their apps and paste them into consumer chatbots. The organization finds out after the fact.
With AgenticliiAgents sit inside the systems your staff already use. People ask about the data they see in their apps, and the agent answers as if it had been pasted into the chat, without it ever leaving your environment.
No security story
IT, legal and compliance cannot approve what they cannot audit. The project dies in review, not in development.
With AgenticliiSeven layers of security and a complete audit trail your reviewers can sign off on.
The blocker is rarely the technology. It is governance. Agenticlii is built to get through review.
The model
One agent. One discipline. One governed data source.
We don't deploy a general-purpose assistant. Each Agenticlii agent is a narrow, accountable specialist for one business function, so its work can be measured, secured and trusted.
You pay an employee to do work, not to answer questions. Our agents work the same way.
Agenticlii agents are built to be employees, not chatbots. They draft the email, update the project plan, prepare the month-end package and flag the exception, all within the permissions you set.
Scoped to a discipline
An agent for finance, another for HR, another for your executive team. Narrow scope means deep competence and output you can measure against a baseline.
Built into your systems
Agents work with the data already in your apps: your SharePoint policies, your reporting model, your templates. No copying and pasting, no spreadsheet someone emailed around, and nothing from the open internet unless you allow it.
Inside your environment
Agents run in your own Microsoft cloud tenant, behind your firewall. Your data stays yours and is never used to train a public model.
Tool-limited by design
An agent can only use the tools you've approved. There is no path to an unsanctioned system, however cleverly it is asked.
LLM-agnostic
Your LLM. Your choice.
Our architecture is LLM-agnostic. Your agents run on whichever large language model you prefer, including one your organization already subscribes to. There is no new AI contract to sign and no lock-in to a single provider.
- Bring your own subscription. Already using an enterprise AI service? Your agents can run on it, under the terms and data protections you have already approved.
- Or let us recommend one. We match each agent to the model that best fits its job, your budget and your compliance needs.
- Switch any time. Change models as prices drop and better ones arrive, without rebuilding your agents.
Azure OpenAIOpenAIAnthropic ClaudeGoogle GeminiOpen-weight modelsMix models by task
Token-optimized
Fewer tokens. Lower cost.
AI providers charge by the token. Our architecture is designed to use as few as possible, which keeps your running costs down.
- Snapshot-first answers. Common questions are answered from pre-built, governed data instead of fresh queries.
- Precise retrieval. Agents send the model only the passages a question needs, not whole documents.
- Right-sized models. Routine requests go to smaller, cheaper models. Complex ones go to the most capable.
- Reuse, not repeat. Repeated answers and instructions are cached instead of being paid for again.
The agents
Specialist AI agents you can add to the team this quarter.
Each agent arrives with a defined role, the Microsoft 365 apps it works in, and a written scope of what it may access. Select an agent to see the work it takes on.
Built for Microsoft 365
AI agents that plug into the Microsoft 365 tenant you already run.
Agenticlii is architected to sit on top of any Microsoft 365 environment. Your people keep working in Outlook and Teams. Your IT team keeps the identity, permissions and policies it already manages.
Approve the connection
A Global Administrator grants consent for the Agenticlii app in Microsoft Entra ID. Sign-in, MFA and conditional access policies carry over unchanged.
Your IT · one sittingPoint each agent at its sources
Choose the SharePoint sites, mailboxes, calendars and reporting models an agent may use. Access follows the permissions and sensitivity labels you have today.
Business owner + IT · short reviewAgents appear where people work
Staff find their agents in Teams chat and Outlook, with no new portal and no new password. Every request is logged from the first day.
Agenticlii · we handle the build
AI security
Seven layers. One environment. No public exposure.
Defence in depth: every layer is built on the assumption that the one outside it may fail. Every request takes the same governed path.
See it work
Follow a request through the architecture
How it works
Every request takes one governed path inside your own cloud tenant. Pick a scenario to watch how each layer handles it.
In short: only your people get in, each agent can only do the job it was given, it sees only the information that job needs, nothing is reachable from the internet, and every action is recorded. Select a layer to see what it means in plain English, with the technical detail for your IT team underneath.
Layer 1 of 7
Identity & authentication
The front door lock
In plain English
Only your own people can use the agents. They sign in with the same work account they use for email, so when someone leaves and IT switches off their account, their access to the agents ends too.
For your IT team
- OAuth 2.0 against your own Microsoft Entra ID, with no anonymous access
- TLS 1.2 or higher on all traffic in transit
- IP restrictions and rate limiting block unknown networks and runaway loops
- Every API call is logged for audit
Why it mattersNothing reaches an agent without a verified identity from your own directory.
No public endpoints
Internal services cannot be reached from the internet.
No raw personal data
Identifying employee and customer detail stays out of scope unless you sanction it.
No credentials in code
Secrets live in a vault and are injected at runtime.
LLM-agnostic, and built on enterprise AI platforms that support HIPAA-compliant configurations and Business Associate Agreements. Hosted entirely within your cloud tenant and behind your firewalls. Your data is never used to train a public model.
The value case
What the business actually gets.
Speed to answer
Questions that took a report request and three days get answered in the meeting where they were asked.
Capacity, not headcount
Analysts, controllers, coordinators and assistants move from lookup work to judgment work.
One version of the truth
Every site and every user gets the same governed number and the same policy answer.
Approvable
A security architecture your IT, legal and compliance reviewers can sign off on. That is what turns a pilot into a deployment.
We build the value case with your numbers before we build the agent. If the payback isn't there, we'll tell you.
Capacity estimator
Illustrative only0hours returned per year
0full-time equivalents
$0capacity value per year
Assumes 46 working weeks and 1,720 hours per FTE. A real value case uses your measured baselines.
Beyond the build
Your people learn the agents. The agents learn your business.
Most AI projects stall after launch because nobody shows staff how to use the tool, and the tool never gets better. Every Agenticlii plan includes hands-on training and an agent that keeps improving, with our team behind it.
Staff training
Training that makes it stick
We teach your people how to get great results from their agents, starting on day one and continuing after launch.
- Role-based launch sessionsLive, hands-on training for each team, built around the questions they actually ask.
- Playbooks for every agentShort guides with example requests, good habits and what each agent can and can't do.
- Internal championsWe train a go-to person in each department to coach colleagues and share what works.
- Office hours and refreshersDrop-in help after launch, plus refresher sessions when agents gain new skills or staff join.
- Adoption reportingSee who is using each agent, for what, and where more coaching would help.
Continuous improvement
Agents that get better every month
An agent's knowledge and skills grow from three sources. Every change is reviewed and approved by the agent's business owner before it takes effect.
Your company's documents
When you publish a new policy, template or report to an approved source, the agent picks it up, so answers stay current without rework.
Your people
Staff rate answers, correct mistakes and suggest better ones right in Teams. Approved corrections become part of what the agent knows.
Our team
Agenticlii specialists review performance every month, tune answers, add new skills and move agents to better or lower-cost models as they arrive.
Staff use the agentQuestions and tasks in Teams and Outlook
Feedback comes inRatings, corrections and new documents
Owner approvesYour business owner reviews every change
Agent improvesBetter answers, new skills, lower cost
The cycle repeats every month. Improvements stay inside your tenant and are never used to train a public model.
How we engage
Four phases. The foundation is built once.
The secured foundation is the expensive part, and you only build it once. Every agent after the first reuses it, so each additional agent costs less than the last.
About 2 weeks
Discovery
Rank candidate use cases by value and feasibility with your business owners. Pick one.
3 to 4 weeks
Foundation
Stand up the secured environment in your tenant: identity, gateway, policy layer, governed data sources and audit.
4 to 6 weeks
First agent
Build, test and validate one agent alongside its business owner, measured against a baseline. Train the team that will use it.
Ongoing
Scale
Add agents for new disciplines on the same foundation. Monthly tuning, new skills and refresher training keep every agent improving.
Pricing
Priced per agent, not per seat.
Every plan includes the seven-layer security foundation, Microsoft 365 integration, a monthly AI token allowance, the language model of your choice (including one you already subscribe to), monitoring and ongoing tuning. Add as many staff users as the plan allows at no extra cost.
Showing US pricing in US dollars.
Launch
Prove the value with one agent.
$1,950USD
per month · 1 agent
25M tokens included monthly
- 1 agent from the catalogue
- Up to 50 staff users
- All seven security layers
- Teams and Outlook access
- Launch training for one team
- Monthly usage and value report
- Business-hours support
Foundation setup: $9,500 one time. Waived on a 12-month term.
Start with LaunchTeam
Most popularCover several functions on one foundation.
$4,950USD
per month · up to 3 agents
75M tokens included pooled
- Up to 3 agents, swappable each quarter
- Up to 250 staff users
- All seven security layers
- Approved actions: drafts, scheduling, Planner updates
- Role-based training and internal champions
- Quarterly value review with your leaders
- Priority support, next business day
Foundation setup: $9,500 one time. Waived on a 12-month term.
Start with TeamEnterprise
Five or more agents across the organization.
from$1,350USD
per agent per month · 5+ agents
25M tokens per agent pooled
- Custom agents built for your roles
- Unlimited staff users
- Audit exports for your compliance tools
- Multi-region and data residency options
- Organization-wide training program
- Named customer success lead
- 99.9% uptime service level
Foundation setup included. Scoped in discovery.
Talk to usDiscovery workshop $2,500
Two weeks to rank your use cases and build the value case. Credited in full when you sign a 12-month plan.
Additional agent $1,450 / mo
Add an agent to Launch or Team at any time on the foundation you already have. Adds 25M tokens a month to your allowance.
Custom agent build from $6,000
A new role that isn't in the catalogue, designed and tested with its business owner.
How token usage is priced
- Plan prices are based on a fixed monthly token allowance, sized for typical business use of each agent.
- Usage above the allowance may result in additional token charges, billed at the rate for the language model your agents use.
- You'll get an alert at 80% of your allowance, so there are no surprises on your invoice.
- If usage keeps running high, we'll recommend ways to bring it down, such as moving routine work to a lower-cost model, or a plan with a larger allowance.
- Bringing your own LLM subscription? Model usage is then billed directly by your provider under your existing agreement, and we adjust your plan price to remove the included token allowance.
Prices in US dollars. Agents can run in US cloud regions for data residency. Annual prices are billed yearly; month-to-month prices can be cancelled with 30 days' notice. Unused tokens do not carry over to the next month. A token is a small unit of text, roughly three-quarters of a word, that AI providers use to measure and bill usage.
Who we are
Built by people who have run the systems these agents plug into.
Agenticlii was founded by a team with deep experience delivering IT, data, analytics and management consulting programs for multi-location organizations across Canada and the United States.
We have sat on both sides of the security review. That is why every agent ships with its scope written down, its access limited to what you approve, and its actions logged where your auditors can see them.
We work with operations, finance and people leaders to pick the job worth doing first, and with IT to make sure the answer to "is this safe?" is yes.
FAQ
Questions reviewers ask first.
Is our data used to train AI models?
No. Agents run inside your own cloud tenant, and your prompts, documents and answers are never used to train a public model.
How much IT time does this take?
Your IT team approves the app in Entra ID, confirms which sources each agent may use, and reviews the security design. We handle the build, the connectors and the testing.
Do agents replace staff or support them?
Both, depending on the role. Some agents absorb the repetitive volume so your team can focus on work that needs judgment. Others cover a function you would otherwise need to hire for. We size that with you during discovery.
Can an agent take actions, or only answer questions?
Agents can draft emails, schedule meetings, update Planner tasks and prepare documents, but only with the tools you approve. Actions that need sign-off can be set to wait for a person.
How is this different from buying Microsoft 365 Copilot licences?
Copilot is a general-purpose assistant licensed per person. Agenticlii agents are specialists licensed per role: each is scoped to one function, grounded in approved sources and governed by the seven security layers. Many organizations run both.
What does an agent cost compared with a hire?
An agent on the Team plan costs about $1,650 a month, well under the loaded cost of most specialist hires. The capacity estimator above shows the hours an agent could return; your value case uses your own numbers.
Do the agents learn from our staff?
Yes, with safeguards. Staff can rate answers and suggest corrections in Teams, and new documents in approved sources are picked up automatically. Your business owner approves changes before they take effect, and nothing your people teach an agent is used to train a public AI model.
Is training included?
Yes. Every plan includes hands-on training for the teams using each agent, playbooks with example requests, and support after launch. Team and Enterprise plans add internal champions and ongoing refresher sessions.
Which AI model do the agents use?
Whichever one you prefer. Our architecture is LLM-agnostic, so agents can run on the language model your organization already subscribes to, such as Azure OpenAI, OpenAI, Anthropic Claude or Google Gemini, or on one we recommend. You can change models later without rebuilding the agent.
Can we use the AI subscription we already pay for?
Yes. If your organization already has an enterprise LLM subscription, your agents can run on it under the terms and data protections you have already approved. Model usage is then billed by your provider, and we remove the included token allowance from your plan price.
What happens if we use more tokens than our plan includes?
Each plan is priced on a fixed monthly token allowance. Usage above it may result in additional token charges at the rate for your selected model. We alert you at 80% of the allowance and help you reduce usage or choose a better-fitting plan.
Where is our data stored?
In your own tenant, in the cloud region you choose, including Canadian and US regions.
What stops someone talking an agent into doing something it shouldn't?
The policy layer. Requests are validated, inputs are sanitized, and the agent can only call pre-approved tools and query patterns. It cannot be talked outside its scope.
The next step
A 45-minute working session.
Bring your IT lead and one business owner. We'll walk through the architecture together and pick your first agent.
Your Microsoft 365 setup and what the agents would need
The discipline with the loudest pain right now
The security questions your reviewers will ask
hello@agenticlii.com