AI-native CRM
that updates itself.
Automated. Human-led. Sovereign.
Built to serve your sales team, not the other way around.
Agents capture every customer conversation and draft the next step from real evidence. Routine work just runs. Anything that matters waits for your call.

Your CRM isn’t empty because your reps are lazy. It’s empty because you hired them to sell, not to type.
A B2B deal moves through calls, mail, messaging apps and meetings, and each step drops a little information. Typing it all back in is manual and draining, so your best sellers are often the ones who update the CRM least. The budget is spent, and the system still delivers little value.
People leave, and the data goes with them
When someone leaves, much of the negotiation context and customer history leaves with them.
Information everywhere
Details are scattered across mail, chat and messaging apps, so managers struggle to see the full picture.
Reports built on gut feeling
Managers open the system and see empty fields. Without real input, the sales forecast rests on gut feeling.
Less data entry. A pipeline you can trust.
Customer records update themselves from every real interaction. Sales stop spending effort on data entry, managers see the true state of the pipeline, and the company keeps its data even when people change.
Connect what you already have.
Mail, calendar and meetings from the sources you choose. Consent, data-subject requests and retention stay with people, never with an agent.
The work nobody in sales wanted to do.
The typing, the stage updates, the chasing and the call notes. Agents do them from the mail, calendar and calls you connected. People check the result.
You never type a contact again.
Contacts, companies and deals appear from the mail, calendar and calls you connected. What an agent wrote stays marked until someone accepts it.
- A contact from a signature, a company from a domain, a deal from a thread
- Duplicates merged with a reason, never silently
- Accept in one tap. Nothing is written until you do
Decide, don't type.
Everything an agent wants to change waits in a deck. Swipe, or use the arrow keys. Accepted changes go to a tray and are sent together, so a rep clears a morning's capture in about a minute. Try it.
Commit is not a feeling.
Every deal in it carries the evidence that put it there: a signed order form in a thread, a date a buyer said out loud in a call. Click a stage and read what stands behind the number.
One scale, one hue stepped by certainty. Figures are illustrative.
Commit
$1,240,000You decide which AI works in the background.
Margince has no model of its own and never resells tokens. Bind each task to a provider you choose, cloud or local, or connect the agent you already run over MCP. You pay the provider directly, with no mark-up.
Pick a provider for each job.
Try it: switch a task to Ollama or vLLM and watch the count change. The real settings screen looks like this, minus the marketing.
Connect over MCP. Or plain REST.
Claude, Copilot Studio, Cursor, or an agent you wrote yourself. It gets Margince's tools, acts with the rights of the person who runs it, and every call is checked against that person's seat and written to the audit log.
Whichever you choose, it works under your rules.
An agent never has more rights than the person it works for, and every call is checked, capped and written down.
Remove someone at 09:00. Their agent stops at 09:00.
The seat and permissions of the person an agent works for are checked on each call, not once at the start.
Every action, named.
Agent, person, and the rule that allowed it. The database refuses any edit or deletion of the log.
No approval lifts it.
An admin sets the number. Agents only draft, people send, and outbound mail always stays inside it.
Most actions just run.
They are logged. A smaller set stops and waits for a person. Which is which follows the rules you set, so nobody keeps that list by hand.
Never approves its own work.
Approvals, consent, data-subject requests and pipeline settings stay closed to agents.
It runs inside your own infrastructure.
On your own servers, in a private cloud, or with a host in the region you choose. There is no vendor-run SaaS. Whether your data stays inside is a configuration you choose, not a promise you have to trust.
Your servers, with Docker
Postgres and Redis. Upgrades are additive only, so a deployed database is never rewritten.
Private cloud
Same installation, your own cloud account. The audit log travels with the database.
A host in your region
For data that has to stay in a given country, with the infrastructure provider you choose.
Where each AI task's text may go
Bind a task to a local model and nothing leaves. The sovereign profile refuses every cloud provider outright.
| Task | Bound to | Text can leave |
|---|---|---|
| Summarise a thread | Claude (cloud) | Yes, the thread |
| Extract contacts from mail | Ollama · local | No |
| Draft a reply | Copilot (cloud) | Yes, the draft context |
| Forecast evidence | vLLM · local | No |
| Approvals, consent, data-subject requests | People only | Never an agent |
The system and its AI models run entirely inside your own infrastructure.
A deployment path for strict data-protection requirements: multinational subsidiaries, banking, securities and insurance. Includes a data impact-assessment packet for your legal team.
Self-contained in your infrastructure
AI models included. No personal data has to leave the boundary your company controls. The exception is messaging channels that run on the provider's own cloud.
A packet for your legal team
Data-protection impact assessment, cross-border transfer documentation and a deployment description stating where data and models run.
An AI layer over your group CRM
Local teams use Margince's AI without giving up the group's standardised CRM. The system stays, the data stays intact.
Starts alongside the CRM you have.
No rip-and-replace. Margince runs next to the CRM you have, and you move the pipeline when your team trusts it. If your old CRM is barely used, start fresh instead.
We migrate your HubSpot data for you.
Your current CRM stays on while your team compares. Gradion moves your HubSpot data into Margince, owners and history included. Switch the pipeline over when Margince is the one people open first. Not before.
Other systems come in by CSV import.
Companies, contacts and leads from any other system come in by CSV import.
Using another system? Tell us when you request a conversation.
Built around your processes.
No metadata engine and no ceiling. A custom field, a workflow or an integration is a proper change in your own copy, protected by types, tests and extension seams that upgrades never touch.
Drop it in, and its presence is the enablement. It only ever touches the public surface, so an upgrade cannot break it.
Whoever adapts it works in the same seams. The tool surface and the web surface are generated from one contract, so what you built for people and what agents use cannot drift apart.
Sub-100 ms interactions, each budget published with its last measurement. The ones nobody has measured yet say so.
1package linecount23import "margince/pkg/fields"45// Production lines a customer runs. Drives the deal-size hint.6var LineCount = fields.Int{7 Key: "line_count",8 On: fields.Company,9 Label: "Production lines",10 Captured: fields.ByAgentWithSource,11}1213func init() { fields.Register(LineCount) }
Run it yourself, or with Gradion at your side.
The full source code, so your IT team can set it up and run it. If you have no dedicated team, Gradion can take on the whole deployment.
Get the full source code
BUSL-1.1 licence; each release converts to Apache 2.0 after 24 months. Security patches are free.
Your IT team installs it
You pay only for acting seats; view seats are free and unlimited. Gradion can set you up with a free test licence first.
You choose where it runs
On your own servers, in a private cloud, or with a host in the region you choose. Gradion does not operate infrastructure for you.
No technical team?
Gradion engineers install it, tailor it to your sales process and support you, under a service package with Gradion directly, with no third party in between.
See service packages →One technology decision. Six leaders who agree.
Putting AI to work on customer data needs sign-off from every key function. Here is what Margince brings to each role: a faster sales team, and company data that stays under control.
Speed up with AI without losing control.
All data stays in the infrastructure you choose.
See the infrastructure →A transparent pipeline, with no more inflated reports.
Every revenue forecast figure comes with clear evidence to check it against.
See the forecast →Free from manual data entry.
Agents keep the CRM up to date and draft follow-up mails. Your team reads, approves and presses send.
See what agents take over →Transparent costs, easy to budget.
View seats are free and unlimited. Acting seats are $29 per month billed annually, or $34 billed monthly. No charges per contact, no mark-up on AI costs.
See pricing →Own your data. Own your customisation source.
Source-available under BUSL-1.1; each release converts to Apache 2.0 after 24 months. Data export at any time.
See customisation →Know where data lives and who accessed it.
Mapped to the data-protection law where you operate, from the GDPR to PDPA and PDPL. You decide where data is stored, and every access is in the audit log. A DPIA packet comes with it.
See security →Engineered in Germany.
Built and backed by Gradion.
Margince is built by Gradion, an independent, self-funded technology company with no pressure from venture investors.
Try it free, or talk to us first.
A free test licence with every feature, run by your IT team on your own data. Questions about deployment or cost? Talk to us for 30 minutes.
Legal team needs documents first? Request the data-protection packetYour details stay confidential.



