
Any proxy, one format, the same AI inventory
If your proxy, secure web gateway or log pipeline can write one JSON object per request, Argent can read it. Argent keeps only visits to AI services it recognizes, records who reached which one, and drops all other lines.
What Argent does with your proxy logs
The same discovery as the [Squid feed](/solutions/integrations/squid-proxy-logs/), for any source that can emit the documented NDJSON schema.
Most of the AI in a company arrives before anyone registers it. Microsoft found 78% of AI users bring their own AI tools to work (Microsoft and LinkedIn, 2024 Work Trend Index). IBM found the share of security incidents involving unapproved ("shadow") AI more than doubled, to 43% from 20%, and those breaches averaged $5.39 million (IBM, Cost of a Data Breach Report 2026). NIST's AI Risk Management Framework, GOVERN 1.6, says "Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities" (NIST AI RMF Playbook).
This is Level 1 of Argent's Discovery: a copy of a log you already keep, with nothing installed and no traffic rerouted.
- What Argent keeps. For each AI service it recognizes, a count of requests and bytes by login, device and client IP, the client family (for example
claude-code/2.0), and what your proxy did: allowed, denied or unknown. - What Argent never keeps. Raw records, URL paths, query strings, user-agent strings, or any host outside its catalog of AI services.
- A signed record. Each rollup is written as one signed discovery record.
- Your richer feed is fine. Fields Argent does not use are ignored, so nobody has to trim an export before sending it.
The schema
One JSON object per line ([NDJSON](https://github.com/ndjson/ndjson-spec)). Argent reads ten fields and ignores the rest.
| Field | What it holds |
|---|---|
ts | Required. Time of the request: an RFC 3339 string or Unix seconds |
user | The login your proxy authenticated, if any |
client_ip | The client's IP address |
device | The device hostname, if your proxy knows it |
host | Required. The destination: a host, host:443 or a URL. Argent keeps only the host |
status | HTTP status your proxy returned |
bytes_out | Bytes sent by the client |
bytes_in | Bytes returned to the client |
user_agent | The client's user agent. Reduced to a family and version, never stored |
action | What your proxy did, such as allow or block. If absent, Argent reads it from status |
At least one of user, device or client_ip must say who made the request. A missing count reads as zero. When a record names more than one identity, Argent groups by the strongest it carries: login, then device, then client IP. The grouping key is a per-tenant digest, not the raw value.
How it connects
Your log pipeline pushes the file to Argent over HTTPS. Argent never reaches into your proxy.
| Direction | Push: your side sends, Argent receives |
| Endpoint | POST /_argent/ingest/proxylog on your Argent gateway |
| Sign-in | A bearer token for the log feed, sent as Authorization: Bearer <token> |
| Format | NDJSON, chosen when you add the feed |
| Compression | Plain, or gzip with Content-Encoding: gzip |
| Tenant | Taken from the token only. Nothing in a record can name a tenant |
Send a batch of lines as the body of one POST. The same error codes apply as for the Squid feed: 401 wrong or rotated token, 403 Discovery off, 413 too large, 429 slow down, 503 retry shortly.
What you configure
Four steps. Nobody installs anything on a laptop.
- Switch Discovery on. It is off until an admin turns it on for the tenant, under Policy > Settings.
- Add a log feed. On the Discovery page, under Sources, add a source of kind Log feed with the format Generic (NDJSON). Argent shows its token once and keeps only a hash; rotate the token to get a new one.
- Map your fields. Have your pipeline write the ten fields above, one object per line, and send the file to
/_argent/ingest/proxylogwith the feed's token. - Check the feed. After the first AI request, the source shows Receiving data under Sources; a feed that goes quiet shows No data received.
Built to take a hostile log line
A log line is data from outside. Argent treats every one as untrusted.
- Only the fields it needs. Every value that reaches a record is checked against a closed character set first. A login outside that set drops the line; it is never rewritten into a different person.
- Secrets never read. A CONNECT target or a URL is cut down to its host. The path, query string and any user name or password inside the URL are discarded unread.
- No user-agent strings kept. The user agent is reduced to a client family and version, such as
claude-code/2.0, so no token, hostname or build path in it reaches the record. - Size and rate limits. Bodies are capped before and after decompression, so a compressed bomb is refused, and each line has its own cap. Each feed has its own rate limit, and one tenant cannot take every slot.
- All or nothing. A push is counted only when the whole body was read within the limits. A refused or retried push leaves nothing behind, so a retry never double counts.
- Silence is not safety. Each feed has its own health counters. A feed that goes quiet reads "No data received", never "no AI usage".
From discovery to control
Discovery is the first half. Control is the second, and the two stay separate, so nothing breaks on day one.
- Review what turned up. Every client Argent sees but nobody registered appears in the Agent inventory tagged Discovered and "Seen via proxy log feed", with types, versions and counts.
- Add managed machines. Level 2 puts the Argent certificate and proxy settings on company machines through your device management: Jamf, Kandji or Microsoft Intune. Only machines from ranges an administrator approved are admitted.
- Register what you keep. Developers register agents with a named human owner, so each gets an identity and a signed record of every request. See Identity Tethering.
- Watch, then enforce. In a new tenant, registered agents start in watch: requests pass unchanged and what each control would have refused or changed goes on the signed record. Then switch controls to enforce, one at a time. See Watch Before You Enforce.
Where this fits
Pages that go deeper.
- Running Squid? Use the Squid log format as it is, with no mapping.
- Capabilities: Shadow AI Discovery, Watch Before You Enforce, One-Click Audit Evidence and all capabilities.
- Goals: See all AI in your company, audit and accountability, compliance evidence and all goals.
- Use cases: securing AI coding agents, rolling out copilots and chat safely and AI in regulated and air-gapped environments.
- Where it runs: air-gapped on-premises, private cloud or your public cloud VPC. Compare them on deployment options.
- Industries: Financial Services, Insurance, Healthcare, Public Sector and all industries.
- More integrations: every system Argent connects to, grouped by type, on the integrations page.
Want this running on your own use cases?
All downloads