Velociraptor: targeted endpoint triage at scale
Use artefacts and VQL collections to answer scoped forensic questions while preserving provenance and limiting unnecessary data.
Scope
For incident responders operating Velociraptor in authorised enterprise environments.
Start with the situation, not the slogan
A security tool should begin with a question. Velociraptor: targeted endpoint triage at scale is presented here as a method for collecting or testing evidence, not as a substitute for an analyst. Decide what fact would change the next decision, then choose the smallest data set and safest operation that can establish that fact.
Tools are confident even when their inputs are incomplete. They will print a result in a reassuringly technical font and leave uncertainty as an exercise for the operator. Record scope, version, configuration, time and source data so the output can be interpreted, repeated and challenged.
Velociraptor can collect rich endpoint evidence quickly. That power requires strong server identity, access control, collection scope, retention, and auditability.
Composite scenario
How this usually reaches the desk
Suppose the immediate question is raised by this clue: Each hunt or collection has an incident, question, endpoint set, artefact version, time range, and owner. Before opening the tool, write the possible explanations and the field or observation that would distinguish them. Collect a known-good comparison where possible. This prevents a colourful result from becoming a conclusion simply because it arrived first.
This scenario combines common operational patterns; it is not presented as a report of one named incident.
What to look for
Begin with preserved, comparable evidence. One signal is rarely proof; use independent observations and a reliable timeline before declaring scope or intent.
Each hunt or collection has an incident, question, endpoint set, artefact version, time range, and owner.
Identify which input creates this observation and whether collection can alter the system. Prefer read-only or passive acquisition when the question permits it.
Server and client trust, administrative actions, exports, and notebook work are auditable.
Check time zone, host, identity, interface and data provenance. A precise result attached to the wrong source remains wrong, only with better punctuation.
Collection cost, endpoint impact, data sensitivity, and retention are understood before execution.
Compare the output with baseline and independent evidence. A match, alert or open port is a lead whose meaning depends on asset purpose, exposure and surrounding activity.
Working examples
Run it, read it, decide what changes
These examples use documentation addresses, test identities and bounded targets. Replace placeholders only inside systems you own or are explicitly authorised to operate. Read the expected result and next action before running the command; a successful command is evidence, not yet a conclusion.
List processes with their executable paths
- Prerequisites
- An enrolled Windows endpoint and authorised collection.
SELECT Pid, Ppid, Name, Exe, CommandLine, Username
FROM pslist()Rows show running processes, parents, paths, command lines and users where available.
A suspicious name is not enough; paths and command lines can be spoofed or absent.
Pivot to the executable hash, signature, parent process and network connections before containment.
Collect current network connections
- Prerequisites
- Authorised live response.
SELECT Pid, Laddr, Lport, Raddr, Rport, Status
FROM netstat()
WHERE Status = 'ESTABLISHED'Established connections are returned with process IDs and endpoints.
A remote address can belong to a CDN or shared service. Absence is only a momentary snapshot.
Join the PID to process data, resolve ownership through approved sources, and compare with proxy, DNS and EDR history.
Find recently created executables
- Prerequisites
- A recorded incident start and a bounded filesystem scope.
SELECT FullPath, Size, Mtime, Btime
FROM glob(globs='C:/Users/*/Downloads/*.{exe,dll,msi,ps1}')
WHERE Mtime > timestamp(epoch=1723968000)Matching files modified after the example epoch are listed.
Timestamps can be changed and the epoch must be replaced with the real UTC start. Other directories and extensions remain out of scope.
Hash selected files through a collection artefact, preserve them according to policy, and correlate with browser and process events.
A practical workflow
Read the whole sequence before starting. Several workstreams may run in parallel, but their evidence, authority and expected outcomes still need to be explicit. Every step below points back to a concrete example; use the example as implementation evidence, not as permission to operate outside the stated scope.
Operational judgement
Operational cost belongs in the design. For Velociraptor, DFIR and Endpoint, estimate collection volume, storage, query time, analyst attention and the impact of credentials used by the tool. A detection that nobody can review is not free; it has simply moved its bill to the incident queue.
Preserve provenance through every transformation. Keep original evidence immutable, note tool and rule versions, and store the exact query or configuration with the result. This is equally useful for incident review, false-positive tuning and the humbling moment when yesterday’s clever filter turns out to have excluded the answer.
Handover
Make the result useful to the next person
Package the useful workflow, not merely the output. Store the question, authority, input source, collection method, tool and rule version, query, time zone, raw result, interpretation and known blind spots. Separate immutable source material from filtered or enriched copies. If the result contains credentials, personal data or confidential content, apply access and retention controls before pasting it into a ticket where it may enjoy a longer life than the system itself.
The operational handover should show how “Secure the server, certificates, administrative access, backups, and audit path before client deployment.” leads to a repeatable decision and how the team verifies “Remove stale hunts and exports, review access, and convert repeated questions into tested procedures.” Add a known test event and an alert for collection failure. State which changes in product version, schema, environment or threat behaviour require review. The result is ready when another authorised analyst can reproduce it, understand its limitations and obtain the same conclusion without borrowing the original operator’s intuition.
Validate before you close
A collection is defensible when a peer can identify who ran it, against which endpoints, with which artefact and parameters, and reproduce the finding.
Capture the test, the expected result and the observed result. Where a person or business owner must accept restored service, name them in the record. A green dashboard can confirm that a component is answering; it cannot confirm that invoices, identities or restored data are trustworthy.
Finish with a compact closure note: the original trigger, confirmed scope, evidence retained, controls changed, tests passed, known gaps, residual risk, and the people responsible for the remaining work. Schedule a review while the timeline is still fresh enough to challenge. The purpose is not to find a person to blame; computers already perform blame with admirable efficiency. The purpose is to make the next response faster, safer and less dependent on one person remembering where the useful log was hidden.
Common mistakes
- Starting with a tool before writing down the question you need to answer.
- Collecting more data than the team can protect, retain, and review.
- Treating an alert or match as a conclusion instead of a lead that needs context.
These errors usually come from haste, unclear ownership or misplaced confidence. Build the safeguard into the runbook: a required evidence field, a second-person review, a rollback test or a specific exit criterion.
Questions people ask when the clock is running
Does a match prove malicious activity?
No. It proves that the input satisfied the rule or query. Confirm provenance, context and related behaviour before assigning intent or impact.
How much data should we collect?
Enough to answer the written question and support likely follow-up, subject to privacy, retention and operational limits. More data is not automatically more truth; it is often more storage with a search box.
What makes the workflow production-ready?
A collection is defensible when a peer can identify who ran it, against which endpoints, with which artefact and parameters, and reproduce the finding. Add ownership, monitoring for collection failure, access control and a documented review cadence.
Safety boundary
Use these steps only on systems you own or are explicitly authorised to assess. Preserve evidence, follow your organisation’s legal and regulatory obligations, and prefer reversible actions when the situation is not yet understood.
Primary references
- Velociraptor Quickstart GuideVelociraptor
- SP 800-61 Rev. 3: Incident Response Recommendations and ConsiderationsNIST
Editorial status: first edition. Review the linked vendor documentation for product- and version-specific changes before acting.