Skip to main content
platform

The only platform that sees what your agents actually do.

Quint reads what an agent says it is about to do, then watches what it actually does. The gap between the two is the signal. It sits underneath every agent at the operating system, so coverage never depends on a vendor shipping a hook.

GET STARTED
Thesis · empiricalQuint · §01

MCP is the 6%.
We see the 94%.

94%
of agent actions in our Claude Code runtime study were native tool_use rather than MCP.
MCP gateways inspect user-configured tools. We intercept every tool call across 7 LLM API parsers and 21+ agent platforms, then cross-check the agent's stated intent against what the kernel actually saw via EndpointSecurity on macOS. Nine competitors were acquired for $2B in the last 12 months; none ship the piece that lives in that gap.
Source · Quint runtime telemetry · Claude Code fleet · 2026illustrative distribution
0 ns
Scoring, on-device · measured
0+
Agent platforms
0
Compliance rules
0
LLMs in the decision path
IThe observation stack

Four layers.
Evaluated in parallel. Cross-validated.

Every tool call is observed from four independent vantage points at once. Risk scoring runs on the machine, with no cloud round-trip in the decision path. Separately, in your cloud tenant, 98 forward-chaining rules map each action to control families: SOC 2 CC6.7 · GDPR Art 32 · HIPAA 164.312 · EU AI Act. All deterministic, zero LLM in the scoring path.

live inspection
14:32:07.441claude-code·Bash: cat config.yml
L1Proxy
Intent
Parsers7 LLM APIs
ProtocolsHTTP · gRPC · MCP
Latency< 2 ms
signaltool_use · Bash"read deployment config"
contributes+12
L2OS Truth
What actually happened
SensorEndpointSecurity
Eventsexec · open · net
KernelES system extension
signalopen(2)/etc/ssh/id_rsa
contributes+38
L3Behavioral
What's normal
Signal families7
Scopes6 (agent → global)
Fingerprint~6 KB
signaldistance4.2σ · far outside envelope
contributes+22
L4Compliance
What the rules say
Control familiesSOC2 · GDPR · HIPAA · ISO
Frameworks16
Rules98 forward-chaining
signalviolatedSOC2 CC6.1 · GDPR Art. 5
contributes+22
composite signal
L1+12L2+38L3+22L4+22
risk0Block · Critical
divergence = (L1 ⊕ L2) amplified by (L3 + L4)pipeline · 47 ms

Nine independent signal families — Bloom-filter novelty, Markov transitions, EWMA temporal drift, Count-Min frequency, HyperLogLog exploration, Jensen-Shannon capability shift, structural flow divergence — all read from one fixed-size ~6 KB per-agent fingerprint whose size is pinned by test regardless of how many actions it absorbs. Sketch-backed, so it never grows unbounded.

IIData governance

What leaves the machine is your call.
Inspect the dial. Audit the wire.

Everything Quint captures lands in your own tenant. In privacy mode, content — prompts, arguments, file bodies — is redacted on-device with a deny-by-default allowlist before a single byte leaves the machine: the cloud sees findings and metadata, never the content. With full capture on, payloads power deeper detection. Either way, the local event log shows exactly what shipped.

plate ii · the wall
on your device
what quint sees
passwordhunter2
emailyou@example.com
credit card4242 4242 4242 4242
ssn123-45-6789
key
stays on your device
password
•••••••#a3f29b
email
••••••••••#6b3ad7
credit card
••••••••••••••••#94f2d1
ssn
•••••••••#2c1e84
Cleartext stops at the wall. Quint's cloud sees the shape of your work, never its body.
IIIBehavioral intelligence

Declared intent.
Then what actually happened.

Other tools pick one signal. Observability watches traces. Gateways watch traffic. Governance writes docs. Quint derives what the agent intended from what it declared, then measures it against what it actually did at the OS level. The divergence between stated intent and observed behavior is the signal, scored against what is normal for this agent at this scope. Intent alone tells you what was promised. OS truth alone tells you what happened. Neither alone tells you: “this support agent is behaving unusually for Alice's queue at 2 am, and what it just did doesn't match what it said it would do.”

divergencedeclared intentobserved behaviorblocked.credential.readsaid./config.yamldid~/.ssh/id_rsa
IVPolicy enforcement

Policy lives on the endpoint.
Evaluated before the action, not after.

Policy pushes to every machine on the next heartbeat and evaluates on-device, ahead of the action — no cloud round-trip on the decision path. Enforcement is observe-first: verdicts are recorded before anything is ever blocked, the same shadow-then-enforce discipline Quint's scope compiler already runs across the fleet, recording what it would have refused. The panel below is illustrative.

Illustrative, not live fleet data
01Say it

Block any agent from reading SSH keys or AWS credentials outside the ~/.aws/sso cache.

Natural language·Compiled in 420ms
02Scope it
Apply to
Agent
Team
Enterprise
match
Read · Bash
target
~/.ssh/** · ~/.aws/credentials
except
~/.aws/sso/
action
BLOCK
severity
critical
Ed25519Signed bundle
03Enforced everywhere
2,847agents
Propagated in 1.2s·100% coverage
Sample rules
  • claude-code
    Read ~/.ssh/id_rsa
    BLOCK
  • devin
    Read ~/.aws/credentials
    BLOCK
  • cursor
    Read ~/.aws/sso/cache
    ALLOW
Enforcement latency · p99 · last 60s
9.0 msSLO 10ms
p50 2.1·max 9.6
VAudit trail

An immutable ledger of every agent action.
Hand it to your auditor. Let them verify it themselves.

Every decision is signed with Ed25519 and SHA-256 chained to the previous block. Any attempt to edit history breaks the chain and breaks the signature, visibly, on the next verification sweep. Export with an OpenSSL one-liner or quint verify.

Chain verified
918,406 blocks·Ed25519 signatures·SHA-256 linkedGenesis → Head
#918,402ALLOW
SHA-256
3b7c0e95
f18d44
14:31:41
claude-code
Bash: docker build
Ed25519·pending
Genesis
#918,403ALLOW
SHA-256
c9e3b701
8a2d66
14:31:52
windsurf
Write: src/utils.ts
Ed25519·pending
#918,404ALLOW
SHA-256
5d2f1a88
c74b13
14:31:58
claude-code
Bash: npm test
Ed25519·pending
#918,405FLAG
SHA-256
a1c8e4f2
d39c56
14:32:07
claude-code
Read: .env.production
Ed25519·pending
#918,406BLOCK
SHA-256
7f3a9b2c
e1f8a0
14:32:11
cursor
Bash: rm -rf node_modules
Ed25519·pending
TamperAny edit breaks the chain
ExportJSONL · Merkle proof · CSV
Verifyopenssl or quint verify
The philosophy
“Every agent action should leave a receipt that an auditor can verify without us, and a verdict your security team can explain without lawyers.
Quint engineering principles · §4
VIOperations

One control plane.
Every machine you have. Every SIEM you use.

A signed macOS package: a daemon plus an Endpoint Security system extension. Linux and Windows sensors, and the browser extension, are on the roadmap. SIEM delivery is configured over the REST API — Splunk, Sentinel, Chronicle, Slack, PagerDuty — with a live event stream for anything that prefers to subscribe.

per-platform interception
macOSshipping
EndpointSecurityexec · fork · open · write syscalls
NETransparentProxysystem extension · MITM TLS
Linuxroadmap
Linux sensoron the roadmap
Windowsroadmap
ETWuser-mode providers · no kernel driver
WFPfilter platform for TLS intercept
Browserroadmap
Chrome MV3declarativeNetRequest
Service workerpersistent agent detection

Your agents are running. See what they're actually doing.

Deploy fleet-wide via MDM. Start with visibility, enforce when ready. No agent configuration required.

Book a demo