Stop updating your CRM.
Aadi perceives the work as it happens, decides what should follow, executes it through your permissions and audit trail, and reports what it can prove. Your team sells; the record keeps itself.
Bring your own AI provider Self-host or managed
Pick a request above.
Aadi plans it, does it, and reports what it can prove — with the record ids to check.
Aadi executes the work and shows its evidence — it does not just answer.
The real interface, running on sample data. Nothing here touches a live workspace.
How it works
A loop, not a database.
Nothing “enters the CRM”. An authorized input arrives and Aadi runs the same five steps over all of them.
01
Perceive
A form submission, an inbound email, an ad lead or an API call — all normalise to one authorized input.
02
Reason
Aadi assembles context from your records, not from the open internet, and plans what should happen.
03
Decide
Scoring, routing and entitlement are deterministic rules — code, not a prompt — so the same inputs always give the same answer.
04
Execute
The platform validates permissions, runs the step through the workflow engine, and writes the audit trail.
05
Verify
Aadi reports what it can prove, with ids. Anything it could not do, it says it could not do.
Note where validation sits: between the decision and any effect. Aadi decides what should happen. The platform decides whether it may.
What you get
Five promises, and nothing we cannot show you.
Each one is a capability in the demo above, not a roadmap item.
A CRM that updates itself
The record is a by-product of doing the work, never a task in itself. Conversations, meetings and outcomes land on the right contact, organization and deal without anyone re-keying them.
Every conversation on one timeline
Email in and out, meetings, notes, documents and every action Aadi took — against one contact, one organization, one deal, regardless of channel or who did it.
Every lead reaches the right person
From form, ad or API to scored, deduplicated, assigned and followed up — with a one-sentence explanation a sales manager can predict and argue with.
Works with the AI you already pay for
Bring your own provider — OpenAI, Anthropic, Google, NVIDIA, open-source or your own endpoint. No captive model, no per-seat AI tax, no lock-in.
AI that executes, not just answers
Aadi searches, reads, drafts, creates and reports with the record ids to check. An assistant that only explains what you could do has failed the job.
Your own AI
Use the model you already pay for.
Models are providers behind an adapter, chosen per workspace with automatic fallback. Aadi is the system around the model — that is the part that does not change when you swap one.
- OpenAI, Anthropic, Google, NVIDIA, open-source, or your own endpoint
- Connect Claude, ChatGPT or Cursor over MCP and operate the CRM from there
- Wide-reaching actions stay behind a human gate, whoever asked for them
{
"aadi": {
"url": "https://www.aadicrm.com/api/mcp",
"headers": {
"x-api-key": "YOUR_KEY"
}
}
}Enterprise trust
The body is the part that can say no.
Aadi reasons and acts. Truth, permissions, validation and audit belong to the platform — and the platform refuses anything that violates them.
Tenant isolation in the database
Postgres row-level security is live and enforcing. The application connects as a restricted role that cannot read another workspace's rows even if the query asks.
Permissions the AI cannot talk its way past
An AI-proposed action runs through exactly the same authorization path as a human one. Confidence may narrow authority; it can never widen it.
Audit as the execution log
Every change carries actor, before/after and source. The audit row and the state change commit together, so one cannot exist without the other.
A defined trust boundary
Aadi reads what you connect, and nothing else. No crawling, no scraping, no background enrichment, no surprise data acquisition.
Your documents stay yours
One attachment service, permissions inherited from the record, and AI that can only open a file the asking user could have opened.
Self-host or managed
The same build runs on your own Docker host or ours. Storage is S3-compatible — AWS, MinIO, R2 or GCS — so nothing pins you to one vendor.
Pricing
Pay for the platform, not for the AI.
Your model costs go to your provider at your rate. We never resell tokens or charge a per-seat AI premium.
Self-hosted
Talk to us
licensed, your infrastructure
Run the whole platform on your own Docker host with your own Postgres and object storage.
- Full CRM and workflow engine
- Bring your own AI provider
- REST API, MCP and webhooks
- Support included
Managed
From £15
per user / month
We run it, patch it and watch it. Your region, your AI provider, your data.
- Everything in self-hosted
- Managed upgrades and monitoring
- Acquisition pipeline and hosted forms
- Priority support
Enterprise
Talk to us
annual
For teams with procurement, a security review and a migration to plan.
- SSO and SCIM on request
- Security review support
- Data-residency options
- Onboarding and migration
FAQ
The questions buyers actually ask.
How is this different from a CRM with an AI assistant bolted on?
The order is reversed. Aadi is the AI system; the CRM is one of its capabilities and the record of what it did. In practice that means the work happens whether or not you open a screen — and when you do, the screen exists to act, not to record.
Which AI models can I use?
Any approved provider, configured per workspace, with automatic fallback: OpenAI, Anthropic, Google, NVIDIA, open-source models, or your own endpoint. Swapping providers changes quality, latency and cost — never business behaviour, contracts or execution semantics.
Can I connect my own AI client?
Yes. Aadi exposes its capability surface over MCP, so Claude, ChatGPT, Cursor or any MCP client can operate the CRM within your permissions. Wide-reaching actions stay behind a human gate on purpose — a bulk send can be drafted by a model and started only by a signed-in person.
Where does my data live?
The self-hosted build runs entirely on your own host, with your own Postgres and your own object storage — we never see it. The managed service runs in one region per workspace, today in AWS Mumbai (ap-south-1); if you need a specific region for a residency or procurement requirement, that is an enterprise conversation rather than a setting, so talk to us before you sign.
What happens to a lead that cannot be assigned?
It escalates with an audited reason. It is never dropped and never left silently unowned — that behaviour is asserted by tests, not by intention.
Is the sandbox on this page real?
It is the real interface rendered from sample data, with no backend attached. We do not connect a public page to a live workspace, and the data you see is invented rather than anonymised.
See it do the work on your own data.
Create a workspace, connect the AI provider you already use, and watch the first lead route itself.