Contact-Ops

In development Does the work of a Relationship manager and contact data steward

We run our company on this. In daily production use at Magic Unicorn.

A relationship manager that earns its autonomy: it merges the duplicates it can prove, asks about the ones it can't, and demotes itself on drift.

Follow the build log See Customer-Ops

Contact-Ops: every person, every connection, one source of truth — Every person. Every connection. One source of truth.

Who it is for

Identity infrastructure the rest of the suite resolves against.

Operators and teams whose customers, partners, vendors, and personal networks are scattered across inboxes, phones, CRMs, and spreadsheets, and who want one clean record per human being.

Why it wins

One person. Every identifier. No duplicates.

One person, one record, everywhere. Customer-Ops and your other apps read the same governed identity instead of hoarding their own copies.

Merges duplicate contacts on its own when an exact email or phone identifier backs the match, and routes name-only lookalikes to your review queue instead of guessing.

Autonomy you can actually trust, because it's earned from real outcomes and revoked automatically, not promised in a demo.

Runs a 5-tier trust ladder: agents earn auto-merge rights from verified outcomes, and a nightly calibration loop demotes any agent whose accuracy drifts.

Nothing is ever lost: every agent action is audited to a tamper-evident ledger, and any merge can be undone for 90 days.

Keeps a canonical, governed registry the whole suite reads: propose-then-approve inbox, hash-chained audit ledger, 90-day reversible unmerge, and 150+ MCP tools for other apps and agents.

How it works

Ingest identifiers. Resolve. Serve the graph.

Feed it your contacts

Import from CSV, CardDAV, or connectors. Everything lands as a proposal, nothing overwrites your data on arrival.

Agents earn their merges

Agents match across email, phone, name phonetics, DOB, address, and company, auto-merging only what they can prove and queueing the rest for you.

You hold the kill-switch

Watch the fleet from the Command Center: live roster, per-agent trust tier, pause and promote controls, and a workspace kill-switch.

See it

Inside the product.

Proof and access

What's proven today.

328duplicates auto-merged, zero failures

99.7%measured merge accuracy

0.020sto score 1,158 candidate pairs

Proof points

328 identifier-backed duplicates auto-merged with zero failures, at 99.7% measured posterior accuracy.
The guardrail works: 84% of high-scoring name-only matches lacked a hard identifier, so every one went to human review instead of auto-merge.
A real 1,448-contact tenant: 1,158 candidate pairs scored in 0.020 seconds.
Every agent action lands in a hash-chained, append-only audit ledger, and merges stay reversible for 90 days.
We run Magic Unicorn's own contact base on it every day, and Customer-Ops agents already upsert real people through it.

Pricing and access

We run our own company's contact base on it every day, and a customer instance is built and staged. Public onboarding isn't open yet; follow the build log and you'll see the door open.

Keep exploring

All products See Customer-Ops

Shipping log

What just shipped.

June 2026: v2.19.0 shipped the operational agent fleet, with the execution bridge, the Agent Command Center console and kill-switch, and the global trust-calibration runner.
June 2026: auto-merge guardrail went live, hard-capping name-only matches to human review.
June 2026: dedup scale fix cut a worst-case 6-hour matching query to milliseconds with indexed phonetic and domain blocking.

FAQ

Plain answers.

Is this a CRM?

No. Contact-Ops owns canonical identity: who a person is, deduplicated and governed. Customer-Ops owns pipeline and deal stages, and it reads Contact-Ops instead of keeping its own copy of people.

How do I know it won't merge two different people?

It only auto-merges when an exact email or phone identifier backs the match. Name-only lookalikes are hard-capped to your review queue, no matter how high they score. In practice that guardrail caught 84% of high-scoring name-only matches that lacked corroboration.

What if an agent gets something wrong?

Every action is written to a hash-chained audit ledger, merges are reversible for 90 days, and the nightly calibration loop demotes any agent whose accuracy drifts. You can also pause an agent or kill the whole fleet from the Command Center.

Can my other tools and agents use the registry?

Yes. Contact-Ops exposes 150+ MCP tools, so other apps and agents propose changes through the same governed inbox with the same audit trail. Customer-Ops federation is already live.

Is my data walled off from other tenants?

Yes. Multi-tenancy is enforced with Postgres row-level security end to end, and even the agent workers run per-tenant under a role that can't bypass it.