Skip to main content

Semgrep Privacy Policy

Semgrep may collect aggregate metrics to help improve the product. This document describes:

Principles

These principles inform our decisions around data collection:

  1. Transparency: Collect and use data in a way that is clearly explained to the user and benefits them
  2. User control: Put users in control of their data at all times
  3. Limited data: Collect what is needed, pseudoanonymize where possible, and delete when no longer necessary

Automatic collection, opt-in, and opt-out

$ semgrep --config=myrule.yaml  # → no metrics (loading rules from local file)
$ semgrep --config=p/python # → metrics enabled (fetching Registry)
$ semgrep login && semgrep ci # → metrics enabled (logged in to semgrep.dev)

Semgrep does not enable metrics when running with only local configuration files or command-line search patterns.

Semgrep does enable metrics if rules are loaded from the Semgrep Registry. This helps maintainers improve the correctness and performance of registry rules.

Metrics may also be configured to be sent on every run, or never sent.

To configure metrics, pass the --metrics option to Semgrep:

  • --metrics auto: (default) metrics are sent whenever rules are pulled from the Semgrep Registry
  • --metrics on: metrics are sent on every Semgrep run
  • --metrics off: metrics are never sent

Alternatively, set the SEMGREP_SEND_METRICS environment variable to auto, on, or off.

Note that certain Semgrep integrators turn on metrics for every run. For example, GitLab's Semgrep SAST analyzer uses --metrics on by default.

Data NOT collected

Data NOT collected ever

We strive to balance our desire to collect data for improving Semgrep with our users' need for privacy and security. After all, we are a security tool! The following never leave your environment and are not sent or shared with anyone.

  • Source code
  • Private rules

Data NOT collected unless explicitly requested

The following data will never leave your environment as part of metrics.

  • Filenames
  • Git commit hashes, messages, authors
  • User-identifiable data about Semgrep’s findings in your code, including finding messages

This data will be sent to Semgrep App only if you explicitly request it, such as with semgrep login && semgrep ci to connect with Semgrep App. Even in that case, your source code and private rules will never be sent.

Data collected as metrics

Semgrep collects data to improve the user experience. Five types of data are collected:

Environmental

Environmental data provide contextual data about Semgrep’s runtime environment, as well as information that helps debug any issues users may be facing; e.g.

  • How long the command took to run
  • The version of Semgrep
  • An anonymous user ID that identifies the machine
  • IP address that triggers a run
  • Value of the CI environment variable, if set
  • Pseudoanonymized hash of the scanned project’s name
  • Pseudoanonymized hash of the rule definitions run
  • Pseduoanonymized hash of the config option. (Note that when a config option downloads a ruleset from the https://semgrep.dev registry, feature usage metrics still include the ruleset name in plain text.)

Performance

Performance data enable understanding of which rules and types of files are slow in the aggregate so r2c can improve the Semgrep program-analysis engine, query optimizer, and debug slow rules; e.g.

  • Runtime duration
  • Total number of rules
  • Total number of files
  • Project size in bytes

Parse Rates

Semgrep reports aggregated parse rate information on a per-language basis; e.g.,

  • Number of targeted files
  • Number of files without any parse-related error
  • Number of bytes across targeted files
  • Number of bytes without any parse-related error

Errors

High-level error and warning classes that Semgrep encounters when run; e.g.

  • Semgrep’s return code
  • The number of errors
  • Compile-time error names, e.g., MaxFileSizeExceeded, SystemOutOfMemory, UnknownFileEncoding

Value

Semgrep reports data that indicate how useful a run is for the end user; e.g.

  • Number of raised findings
  • Number of ignored findings
  • Pseudoanonymized hashes of the rule definitions that yield findings
  • The Semgrep features used during the scan

Pseudoanonymization

Certain identifying data (e.g. project URLs) are pseudoanonymized before being sent to the r2c backend.

"Pseudoanonymized" means the data are transformed using a deterministic cryptographically secure hash. When the input data are unknown, this hash is expensive to reverse. However, when input data are known, a reverse dictionary of identifiers to hashes can be built. Hence, data are anonymous only when the source values are unknown.

We use a deterministic hash to:

  • Track performance and value improvements over successive runs on projects
  • Remove test data from our metrics

Using a deterministic hash, however, implies:

  • An entity that independently knows the value of an input datum AND who has access to r2c's metrics data could access metrics for that known datum

r2c will:

  • Treat collected metrics data as secret, using application-security best practices, including (but not limited to)
    • Encryption during transit and rest
    • Strict access control to data-storage systems
    • Application-security-policy requirements for third parties (e.g. cloud-service providers; see "data sharing" below)
  • Only correlate hashed data to input data when these inputs are already known to r2c (e.g. publicly available project URLs for open-source projects, or projects that log in to the Semgrep Registry)

Description of metrics fields

CategoryFieldDescriptionUse CaseExample DatumType
Environment
Timestamps (started/sent)Time when the event firedUnderstanding tool usage over time2021-05-10T21:05:06+00:00String
Event IDA random UUID generated when sending the event.Deduplicating events in case of issues during transmission222bcccd-9dc2-4d10-ac3a-5692460e77eeString
Anonymous User IDA random UUID generated on first run.Unique users per ruleset and feature. Understanding percentage of logged in users.5f52484c-3f82-4779-9353-b29bbd3193b6String
VersionSemgrep version being usedReproduce and debug issues with specific versions0.51.0String
Project hashOne-way hash of the project URLUnderstand performance and accuracy improvementsc65437265631ab2566802d4d90797b27fbe0f608dceeb9451b979d1671c4bc1aString
Rules hashOne-way hash of the rule definitionsUnderstand performance improvementsb03e452f389e5a86e56426c735afef13686b3e396499fc3c42561f36f6281c43String
Config hashOne-way hash of the config argumentUnderstand performance and accuracy improvementsede96c41b57de3e857090fb3c486e69ad8efae3267bac4ac5fbc19dde7161094String
Is authenticatedWhether the user logged in to semgrep.dev with semgrep loginUnderstand popularity of logged in featuresfalseBoolean
CINotes if Semgrep is running in CI and the name of the providerReproduce and debug issues with specific CI providersGitLabCI v0.13.12String
Client IPIP address that triggered a runUnderstand broad ruleset usage0.0.0.0String
Performance
DurationHow long the command took to runUnderstanding aggregate performance improvements and regressions14.13Number
Total RulesCount of rulesUnderstand how duration is affected by #rules137Number
Total FilesCount of filesUnderstand how duration is affected by #files4378Number
Total BytesSummation of target file sizeUnderstand how duration is related to total size of all target files40838239Number
Rule StatsPerformance statistics (w/ rule hashes) for slowest rulesDebug rule performance[{"ruleHash": "7c43c962dfdbc52882f80021e4d0ef2396e6a950867e81e5f61e68390ee9e166","parseTime": 0,"matchTime": 0.05480456352233887,"runTime": 0.20836973190307617,"bytesScanned": 0}]StatsClass[]
File StatsPerformance statistics for slowest filesDebug rule performance[{"size": 6725,"numTimesScanned": 147,"parseTime": 0.013289928436279297,"matchTime": 0.05480456352233887,"runTime": 0.20836973190307617}]StatsClass[]
Parsing
Total FilesCount of files, on a per-language basisUnderstand parsing performance143Number
Total BytesSummation of target file size, likewise groupedUnderstand parsing performance41244Number
Parsed FilesCount of files without parse errorsUnderstand parsing performance140Number
Parsed BytesCount of bytes without any parse errorsUnderstand parsing performance40312Number
Errors
Exit CodeNumeric exit codeDebug commonly occurring issues and aggregate error counts1Number
Number of ErrorsCount of errorsUnderstanding avg #errors2Number
Number of WarningsCount of warningsUnderstanding avg #warnings1Number
ErrorsArray of Error Classes (compile-time-constant)Understand most common errors users encounter["UnknownLanguage", "MaxFileSizeExceeded"] ErrorClass[]
WarningsArray of Warning Classes (compile-time-constant)Understand most common warnings users encounter["TimeoutExceeded"]WarningClass[]
Value
Features usedList of strings that identify Semgrep features usedUnderstand what features users find valuable, and what we could deprecate["language/python", "option/deep", "option/no-git-ignore", "key/metavariable-comparison"]Object
Rule hashes with findingsMap of rule hashes to number of findingsUnderstand which rules are providing value to the user; diagnose high false-positive rates{"7c43c962dfdbc52882f80021e4d0ef2396e6a950867e81e5f61e68390ee9e166": 4}Object
Total FindingsCount of all findingsUnderstand if rules are super noisy for the user7Number
Total NosemsCount of all nosem annotations that tell semgrep to ignore a findingUnderstand if rules are super noisy for the user3Number

Anonymous user ID

anonymous_user_id: "5f52484c-3f82-4779-9353-b29bbd3193b6"

This ID is stored in the ~/.semgrep/settings.yml file. If the ID disappears, the next run will generate a new one randomly. The Semgrep team uses this to answer the following questions:

  • How many people use a given rule/ruleset/snippet?

    This helps our security researchers assess their performance, and we're planning to make these numbers public for all rule authors in the community.

  • What percentage of users log in?"

    We use this to evaluate our success as we build new authenticated features on Semgrep App.

This ID will only ever be sent with Semgrep's metrics collection endpoint, meaning it cannot be used to track users across the web.

Feature usage

"features": ["language/python", "option/deep", "option/no-git-ignore", "key/metavariable-comparison"]

Examples of such features are: languages scanned, CLI options passed, keys used in rules, or certain code paths reached, such as using an :include instruction in a .semgrepignore file. These strings do NOT include user data or specific settings. As an example, for semgrep scan --output=secret.txt Semgrep sends "option/output" but will NOT send "option/output=secret.txt".

The list of features tracked as of June 2022 is:

  • language: What languages were scanned
  • cli-flag/cli-envvar: What options were configured (does NOT include their value)
  • config: What method was used to retrieve rules (does NOT include any of the rule)
  • registry-query: The value of a --config r/foo.bar.baz setting, limited to the first word (e.g. r/foo.. in this example)
  • ruleset: The value of a --config p/foobar setting
  • semgrepignore: Whether an :include statement was used in a .semgrepignore file
  • subcommand: What subcommand was used (e.g. scan or ci)

The Semgrep team uses this to answer the following questions:

  • How many people use a given feature?

    This guides our development, and lets us decide when and how to deprecate features.

  • How does feature usage affect finding counts, error counts, and performance?

    We use this to evaluate experimental features and understand their production-readiness.

Sample metrics

This is a sample blob of the aggregate metrics described above:

{
"started_at": "2021-05-10T21:05:06+00:00",
"sent_at": "2021-05-10T21:05:09+00:00",
"event_id": "222bcccd-9dc2-4d10-ac3a-5692460e77ee",
"anonymous_user_id": "5f52484c-3f82-4779-9353-b29bbd3193b6",
"environment": {
"version": "0.51.0",
"ci": "true",
"configNamesHash": "ede96c41b57de3e857090fb3c486e69ad8efae3267bac4ac5fbc19dde7161094",
"projectHash": "c65437265631ab2566802d4d90797b27fbe0f608dceeb9451b979d1671c4bc1a",
"rulesHash": "b03e452f389e5a86e56426c735afef13686b3e396499fc3c42561f36f6281c43",
"isAuthenticated": false
},
"performance": {
"runTime": 37.1234233823,
"numRules": 2,
"numTargets": 573,
"totalBytesScanned": 33938923,
"ruleStats": [{
"ruleHash": "7c43c962dfdbc52882f80021e4d0ef2396e6a950867e81e5f61e68390ee9e166",
"parseTime": 0,
"matchTime": 0.05480456352233887,
"runTime": 0.20836973190307617,
"bytesScanned": 0
}],
"fileStats": [{
"size": 6725,
"numTimesScanned": 147,
"parseTime": 0.013289928436279297,
"matchTime": 0.05480456352233887,
"runTime": 0.20836973190307617
}]
},
"parse_rate": {
"python": {
"num_targets": 102,
"targets_parsed": 101,
"num_bytes": 985123,
"bytes_parsed": 993419
},
"ruby": {
"num_targets": 12,
"targets_parsed": 12,
"num_bytes": 341027,
"bytes_parsed": 341027
}
},
"errors": {
"returnCode": 1,
"errors": ["UnknownLanguage"],
"warnings": ["MaxFileSizeExceeded", "TimeoutExceeded"]
},
"value": {
"ruleHashesWithFindings": {"7c43c962dfdbc52882f80021e4d0ef2396e6a950867e81e5f61e68390ee9e166": 4},
"numFindings": 7,
"numIgnored": 3,
"features": ["language/python", "option/deep", "option/no-git-ignore", "key/metavariable-comparison"]
}
}

Data collected when explicitly requested

For Semgrep App users running semgrep ci while logged in, data is sent to power your dashboard, notification, and finding management features. These data are ONLY sent when using semgrep ci in an App-connected mode and are not sent when not logged in.

Two types of data are sent to r2c servers for this logged-in use case: scan data and findings data.

Scan data provide information on the environment and performance of Semgrep. They power dashboards, identify anomalies with the product, and are needed for billing. The classes of scan data are:

  • Project identity (e.g., name, URL)
  • Scan environment (e.g., CI provider, OS)
  • Author identity (e.g., committer email)
  • Commit metadata (e.g., commit hash)
  • Review and review-requester identifying data (e.g., pull-request ID, branch, merge base, request author)
  • Scan metadata, including type of scan and scan parameters (e.g., paths scanned and extensions of ignored files)
  • Timing metrics (e.g., time taken to scan per-rule and per-path)
  • Parse metrics (e.g., number of files targeted and parsed per-language)
  • Semgrep environment (e.g., version, interpreter, timestamp)

Findings data are used to provide human readable content for notifications and integrations, as well tracking results as new, fixed, or duplicate. The classes of findings data are:

  • Check ID and metadata (as defined in the rule definition; e.g., OWASP category, message, severity)
  • Code location, including file path, that triggered findings
  • A one-way hash of a unique code identifier that includes the triggering code content
  • Dependency name and version (only sent when using Semgrep Supply Chain or the --sca flag)
  • Source code is NOT collected

Registry fetches

Certain Registry resources require log-in to the Semgrep Registry. Log in may be performed using your project URL, or a Semgrep.dev API token. When using these resources, your project's identity will be recorded by the Semgrep Registry servers.

Data sharing

We use some third party companies and services to help administer and provide Semgrep, for example for hosting, customer support, product usage analytics, and database management. These third parties are permitted to handle data only to perform these tasks in a manner consistent with this document and are obligated not to disclose or use it for any other purpose.

We do not share or sell the information provided to us with other organizations without explicit consent, except as described in this document.