JSON output

The --json output is valid JSON with no ANSI codes or decorative text on stdout, so you can pipe it straight to files, jq, CI steps, IDE extensions, MCP adapters, or AI agents.

php artisan telescope:inspect --requests --queries --exceptions --jobs --last=1h --json > telescope.json

Envelope

{
    "schema_version": "1.0",
    "command": "telescope:inspect",
    "generated_at": "2026-08-22T12:00:00.000000Z",
    "agent": null,
    "filters": {
        "types": ["request", "query"],
        "last": "1h",
        "from": null,
        "to": null,
        "resolved_window_utc": { "from": "...", "to": null },
        "limit": 50,
        "min_duration_ms": null,
        "route": null,
        "methods": [],
        "statuses": [],
        "failed_jobs_only": false,
        "connection": null,
        "search": null,
        "batch_id": null,
        "full": false,
        "fail_on": []
    },
    "summary": { "total_entries_in_window": 184 },
    "violations": [],
    "items": []
}

(Abridged example; agent, summary.scan and the other keys below are always emitted.)

  • schema_version: bump-major contract. Within 1.x, existing fields keep their meaning; new fields may appear. A formal JSON Schema lives at schema/telescope-inspect-v1.schema.json.
  • generated_at: UTC ISO 8601.
  • agent: name of the detected AI agent when JSON was auto-selected for it, otherwise null.
  • filters: exactly what you asked for plus the resolved UTC window. batch_id is null unless --batch is used. --show is intentionally not echoed: it selects an entry directly.
  • summary.total_entries_in_window: exact count of Telescope entries in the window across all types (not just the selected flags). In --batch mode it is the number of entries in that batch.
  • summary.entries_by_type: per-type counts (only types with entries).
  • summary.items_returned: how many items were included per type (--limit caps the combined listing, newest first).
  • summary.analysis: analyzer results for requests, queries, exceptions and jobs when those types are selected (details). Each summary carries *_analyzed and rows_scanned counters so partial coverage is visible.
  • summary.content_filters_present: whether any content filter (--min-duration, --route, --method, --status, --failed, --connection, --search) was supplied. Analysis always covers the whole window regardless of these filters.
  • summary.scan: { "limit", "truncated" }, the bounded newest-first scan ceiling and whether it was hit while more entries were requested. truncated: false guarantees the whole window for the selected types was examined; analyzer walks expose their own coverage through rows_scanned.
  • violations: --fail-on checks that matched (empty array otherwise); mirrors exit code 3.

For --show=<uuid> the envelope contains a single "entry" object instead of "items" and has no violations key. A missing UUID exits 1 in every mode: humans see a styled error, machine modes get the message on stderr with an empty stdout.

Items

Each item is one normalized entry. Common keys:

Key Always present
uuid yes
type yes, the Telescope entry type string
batch_id nullable
created_at nullable, UTC ISO 8601

Type-specific fields are merged at the same level. Anything ending in _ms is milliseconds.

Example: query

{
    "sql": "select * from `orders` where `user_id` = ?",
    "connection": "mysql",
    "driver": "mysql",
    "duration_ms": 12.34,
    "slow": false,
    "file": "/app/app/Models/Order.php",
    "line": 55,
    "query_hash": "9f8a...",
    "uuid": "...",
    "type": "query",
    "batch_id": "...",
    "created_at": "2026-08-22T11:59:59.000000Z"
}

Field reference per type

Type Fields + sensitive (only with --full)
request method, uri (path only), controller_action, middleware[], response_status, duration_ms, memory_mb, ip_address query_string, headers, payload, session, response, response_headers
client_request method, uri (path only), response_status, duration_ms query_string, headers, payload, response, response_headers
query sql, connection, driver, duration_ms, slow, file, line, query_hash bindings
exception class, message, file, line, occurrences context, trace, line_preview
job name, queue, connection, status (pending/processed/failed), tries, timeout, exception_message data
command command, exit_code arguments, options
schedule command, description, expression, timezone, user output
cache operation (hit/missed/set/forget), key, expiration value
dump entry_point_type, entry_point_uuid dump
event name, listeners[], broadcast payload
gate ability, result, message, file, line arguments
log level, message context
mail mailable, subject, queued, from[], to[], cc[], bcc[], reply_to[] html, raw
model action, model, count changes
notification notification, channel, queued, notifiable response
redis command (verb and key), connection, duration_ms arguments
view name, path, shared_keys[], composers[] none
batch name, total_jobs, pending_jobs, failed_jobs, processed_jobs, progress, queue, connection, allows_failures, cancelled_at, finished_at options

Analysis summaries cap their output: top 10 routes / query patterns / signatures / recurring failures, top 5 N+1 candidates and latest exceptions. summary.analysis.*.rows_scanned tells you exactly how much of the window fed them.

Under default redaction the sensitive fields listed above are removed from items entirely (the keys are absent), not set to null. When --full is enabled the same fields appear; a field Telescope never recorded is then present as null. Consumers can therefore distinguish "intentionally omitted" from "not recorded".

Nullable non-sensitive fields may be omitted values (null) rather than missing keys; consumers should tolerate both.

Sensitive values

Fields such as payload, headers, session, query_string, bindings, trace, context, data, value, dump, and output are redacted by default: they are removed from JSON entirely. Pass --full (or set redact_sensitive: false in config) to include them, but only do this on machines where that data is safe to expose.

Note two deliberate residual exposures: request URIs keep their path (query strings are gated) but free-text fields like SQL statements, log/exception messages, mail subjects, addresses, and cache keys are always visible - they are the point of those entries.

All string values are truncated at value_limit characters (default 1000) so a single huge payload can never blow up a consumer.

NDJSON

--ndjson emits one compact JSON item per line, ideal for streaming into jq:

php artisan telescope:inspect --queries --last=1h --ndjson | jq 'select(.duration_ms > 500)'

Requires at least one type flag, --batch=<id>, or --show=<uuid>. Combine with --full as needed. Violation notes go to stderr so piped lines stay parseable.

Exit codes

Code Meaning
0 success (including valid empty results)
1 runtime failure: storage missing, UUID or batch id not found
2 invalid usage: bad values or impossible combinations
3 issues found via --fail-on

In JSON and NDJSON modes stdout carries data only; diagnostics are plain text on stderr, so 2>&1 is safe but never required.