KELS lets apps and AI agents read MongoDB through reviewed, signed queries.
For teams on MongoDB. Your data never leaves your network.
TENANT_DENIED"month": {"$ne": null}422 TYPE_MISMATCHAPPROVAL_REQUIREDExample requests and sample data. Pick one and see what the agent sends and what KELS answers.
{
"query_id": "visits-by-week", "query_version": 3,
"data_tenant_id": "clinic-0042",
"parameters": {
"month": { "type": "date", "value": "2026-10-01T00:00:00Z" } }
}
{
"status": "ok",
"data": { "row_count": 4, "rows": [
{ "week": "2026-10-05", "visits": 41 },
{ "week": "2026-10-12", "visits": 37 }, …
] },
"decision": { "outcome": "ALLOW" }
}
ALLOWThe query was reviewed and published earlier. KELS checks the signature, adds this clinic’s filter itself and returns only the fields the query was approved for.
Shortened for reading: real query IDs are UUIDs, and requests also carry a purpose code.
To let an agent answer questions from your database, most teams hand it a connection string. From then on it can read anything that user can read, for any customer, and nobody sees which queries it runs.
| The question | With a connection string | With KELS |
|---|---|---|
| What can it run? | Any query it writes, including ones nobody has looked at | Only queries a person reviewed, tested and published, called by ID |
| Whose records? | Every customer in the database | Only the customer it was granted; KELS adds that filter itself |
| Passwords, tokens, ID numbers? | Returned if the query asks for them | Restricted fields can never be queried; sensitive ones can need an approval |
| Where do credentials live? | In the agent’s prompt, config or tools | Only in the KELS agent inside your network; the AI agent holds an API key you can revoke |
| What can you see afterwards? | Database logs, if they were on | Every run with who, which query, which customer and why, in a tamper-evident audit trail |
The KELS agent runs next to your database. It starts every connection to the KELS cloud; nothing in the cloud can reach into your network.
The portal where your team classifies fields, reviews and publishes queries, approves requests and reads the audit trail.
Calls a published query by its ID, with typed values and an API key.
Checks the signature, the key, the customer and every value, then runs the query and filters the result.
Unchanged. KELS only needs one user with the read role.
Your team sets it up once per question. After that, the app or agent calls it like any other API.
Install the KELS agent on a small Linux machine next to MongoDB with one package and a guided setup. An admin approves each database in the portal.
Done once, by your platform team
sudo apt install ./kels-agent_0.3.1-1_amd64.deb sudo kels-agent setup --cloud <the address the portal shows>
KELS lists every field and suggests a level from what it sees: emails, phone numbers, card numbers, Aadhaar and Social Security numbers. You confirm with one click. Fields nobody has classified are never returned.
Done by whoever owns the data
| Field | Level | Why |
|---|---|---|
visit_date | Ordinary | Date |
phone | Sensitive | Looks like a phone number |
ssn | Restricted | Social Security number shape |
A query is drafted in the portal, run against staging, and published by a different person from its author. KELS gives it an ID and a version, signs it, and writes a plain-English summary of exactly what it returns.
Done by an engineer and a reviewer
The portal hands you ready-to-run code. It sends the query ID, the customer and typed values to your KELS agent. Values can only ever be values, so they cannot change what the query does.
Done by your app, an AI agent, or a script
curl -X POST "$KELS_AGENT_URL/v1/execute" \
-H "X-Kels-Api-Key: $KELS_API_KEY" \
-H "Idempotency-Key: $(uuidgen)" \
-H "Content-Type: application/json" \
-d '{"query_id": "…", "query_version": 3,
"data_tenant_id": "clinic-0042",
"purpose_code": "care_planning",
"parameters": {"month": {"type": "date",
"value": "2026-10-01T00:00:00Z"}}}'
Also in Python, Java, Kotlin, Go, C#, Node.js and Rust. The portal writes the code for each query. No KELS SDK to install.
What ships today in KELS 0.3.1, grouped by when it does its work.
Teams on MongoDB who hold customer data and want to put it in front of apps or AI without handing out the keys.
MongoDB ended the Atlas Data API and custom HTTPS endpoints in September 2025. KELS gives you HTTP calls by query ID, with no database credentials in the caller.
Give an assistant a set of safe questions it can ask. Each answer stays inside one customer’s records, and sensitive fields need a person’s approval.
Every query is reviewed, signed and audited, and customer data never leaves your network. KELS helps you meet HIPAA and SOC 2 access controls; your own audit still decides compliance.
Planned for KELS 1.0. None of this ships yet, and we will not sell it before it does.
AI agents discover and call your published queries as tools, directly.
Your AI drafts the query. Your rules decide whether it runs, and sensitive data still needs a person.
Queries across collections and inside lists.
After 1.0.
Anything else: write to us at hello@kels.io.
No. Documents, results and the values in requests stay in your network. The cloud sees field names, types and run records so your team can review and audit.
Your database stays as it is; KELS only needs one read-only user. Your app replaces its database query with an HTTP call to the KELS agent.
One package on a Linux machine near the database (x86_64; Ubuntu, Debian, RHEL, Rocky, Alma or Amazon Linux). We suggest 2 CPUs and 2 GB of memory to start.
MongoDB replica sets, self-hosted or Atlas. KELS is tested on MongoDB 8.2.
Not today. Agents call queries your team has published. From KELS 1.0 an AI can draft queries, but your rules and a person still decide what runs.
No product is compliant on its own. KELS helps you meet HIPAA and SOC 2 access controls with reviewed queries, least privilege and a full audit trail.
Pilots are arranged with each team. Talk to us and we will work out what fits.
We are taking a small number of design partners. We help you set it up, and you get these limits for the pilot:
hello@kels.io