Requires the write scope or higher. Adds a diagnosis note to an incident; it appears on the incident's page in order, stored with author 'agent' (no author argument exists). Notes are append-only: no edit or delete exists, so a correction is a new note and the history stays evidence. Several notes per incident are normal, up to a cap of 50, and a closed incident still accepts them. A note can record a failure that was not fixed as well as one that was. Without a note, the incident page shows only the alert.

Requires the write scope or higher. Counts a discovered trace source's traces under an agent from now on. Without agent_id, a new agent named after the source is created (409 AGENT_EXISTS when its slug is taken). With agent_id (e.g. a suggested_agent_id), the source merges into that agent. Earlier traces stay where they are (backfilled is false). Re-adopting into the same agent changes nothing; into a different one is 409 ALREADY_ADOPTED. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the admin scope or higher. Creates an API key. The plaintext key appears only in this result and cannot be retrieved again. expires_at makes a short-lived key. create_ingest_key, which needs only the write scope, mints a single-monitor tracing key.

Requires the write scope or higher. Creates a notification destination (channel) that monitors can route alerts to, from the fields of the chosen kind; unrelated fields are ignored. Non-email kinds are usable immediately; an email destination starts unverified, sends a confirmation link, and can join a route once the link is clicked. A project holds at most 25 destinations (DESTINATION_CAP_REACHED beyond that). Returns the new channel id, which set_route takes.

Requires the write scope or higher. Creates a tracing key: an ingest-scoped key bound to one monitor, able to send traces, metrics, logs and pings for it and nothing else; every REST call refuses it. For automation that stores the key itself (a CI secret, a deployment's secret store). The plaintext key appears only in this result, which puts it in the conversation; the console's Create a tracing key keeps it out of chat.

Requires the write scope or higher. Creates a monitor, or updates the existing one when slug matches (an upsert; the result says 'updated'). Heartbeat and ci monitors take schedule_kind: 'simple' with period_s, 'cron' with cron_expr, or 'on_demand' with neither. http monitors take probe_url and probe_interval_s; probe_expected_status and probe_expected_body define a healthy response, and a probe with neither only checks that something answered. ci_provider can be set only here, and the webhook secret it returns appears only in this result.

Requires the write scope or higher. Creates a status page: one page showing the current status and recent history of a chosen set of monitors, for people who cannot log in to the project. Pages are private unless visibility is 'public', which publishes every monitor name on the page.

Requires the write scope or higher. Declares, at the start of a run, the criteria its success ping body is judged by when the run closes, so the run does not grade itself: a success whose body fails any declared criterion is recorded as a failed run with cause 'assertion', whatever the exit code. One declaration per rid, immutable (a second call is a conflict). For use right after the run's /start ping; get_ping_instructions' expectations_how_to has a worked example.

Requires the write scope or higher. Permanently deletes an agent from the registry by UUID; cannot be undone. Its monitors are not deleted: each survives with its ping history and incidents, becomes unowned (agent_id null) and keeps running. update_monitor can attach a survivor to another agent, and delete_monitor removes a monitor.

Requires the write scope or higher. Permanently deletes a notification destination (channel); cannot be undone. It is removed from every monitor's routing, so an event type routed only to it stops notifying anyone, with no incident to show for it. get_monitor's routes shows which event types route to it, and set_route changes routing without deleting the destination.

Requires the write scope or higher. Permanently deletes a monitor by UUID; this cannot be undone. Its pings, run steps, traces, incidents, routes, alert templates, assertions and delivery history are deleted with it, and the ingest keys bound to it (from create_ingest_key) are deleted too, so they stop working.

Requires the write scope or higher. Stops routing one event type of a monitor to any destination, so its alerts for that event go nowhere; every other event type's routing is unchanged. set_route with the remaining ids drops a single destination instead. An event type with no routing answers "route not found".

Requires the write scope or higher. Permanently deletes a status page; cannot be undone, and its public URL stops working immediately. The monitors on it are unaffected and keep running and alerting. update_status_page with visibility 'private' stops sharing without deleting.

Requires the write scope or higher. Turns a scan of a repository or host (sources, every scheduled job found) into monitors. Creates one monitor per source not already monitored; never deletes, pauses or edits existing monitors, so a scheduled re-run is safe drift detection. Returns created, existing (unmodified, hand-tuned values intact) and orphaned (source not in this scan; still running and alerting). Each created monitor can page a person; this endpoint has no delete path. source_kind plus source_ref is the diff key, so source_ref has to stay stable: a changed ref creates a duplicate; a crontab ref names the file and the command, so each line is its own source. Kinds: 'crontab' (crontab -l, /etc/cron.d/*, /etc/crontab), 'github-actions' (on.schedule.cron), 'k8s-cronjob' (spec.schedule), 'systemd-timer' (OnCalendar=). A source without schedule_cron becomes on-demand. crontab and systemd-timer fire in the host's local time (timedatectl show -p Timezone --value or readlink /etc/localtime prints it); github-actions and k8s-cronjob in UTC. A stated tz is stored as given: the API cannot tell a guessed 'UTC' from one read from the host. list_monitors shows each source.

Requires the read scope or higher. Exports existing monitors, destinations, routes, alert templates and status pages as Terraform HCL, with import blocks so they are adopted rather than recreated. Secrets are not exported; the output references Terraform variables that hold them.

Requires the read scope or higher. Gets one agent by UUID, with the same fields as list_agents: its live status rollup, usage_24h and top_dependencies. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. What one agent calls, heaviest first, from its traces: each model, tool, HTTP host, database, queue, RPC endpoint or agent: calls, errors, error_rate (0-1), p50_ms, p95_ms (bucket ceiling; p95_is_floor: over 60s), daily series, and for a model tokens, cost_usd and cost_source (client, estimated or mixed). operations: up to five span names per outgoing row, sampled from the ten newest traced runs. At most 50 rows; more counts the rest. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Model usage, one row per model per UTC day: tokens_in (cache reads included, so tokens_cache_read is not added to it), tokens_out, tokens_cache_read, tokens_cache_write, cost_usd (decimal text) and cost_source: client (the tool reported its cost), estimated (LastPing priced tokens at API list prices), mixed (a traced day holds both) or empty (unknown). origin is traces or metrics; a day and model can have one of each, never summed. With id, one agent's usage; without, the project's, traces only, plus by_agent (costliest first). Both carry by_project: per project (a Claude Code session's folder) its runs, tokens and cost from traced runs only (project_scope traces_only). by_project sums each run's traced totals, while one agent's days prefer a Claude Code metrics export's reported cost, so the two need not match. With project, days become one row per UTC day with model and provider empty; project narrows days and by_project, not by_agent. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Gets a monitor's custom alert message templates: a map of event-type (or event-type/cause) keys to template strings. Keys: 'down', 'recovery', 'fail', 'every-run', 'success', 'started', 'blocked', 'note', or 'event_type/cause' (e.g. 'down/silence'). An empty result means every alert uses the built-in plain-language defaults.

Requires the read scope or higher. Gets one incident and its ordered timeline: run_started, step, run_failed/cancelled/blocked, incident_opened, alert_delivered/failed/suppressed/pending (destination and attempts; down and fail alerts only), note, incident_resolved. For what the run was doing, who was paged and what was tried. run_* events are matched on the run id recorded at opening, so none means no run was recorded. Delivery error text is omitted. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Gets one monitor by UUID with its full configuration, including assertions (conditions a successful run's ping body has to satisfy), guards (ceilings on a number the job reports) and routes (which destinations receive which event type); each is absent when empty. update_monitor's assertions and guards and set_route each replace a whole set, and this result holds the current sets.

Requires the read scope or higher. Returns what a monitor needs in order to report: its ping URLs, copy-paste snippets (curl_success, curl_start, curl_fail, curl_step, run_example) and three reporting mechanisms, for wiring up a new monitor. reporting_options holds the rule for choosing between them: how_to is the manual protocol and works in any agent with no prerequisite (with expect_every_s, a lapse opens an incident); hook_install is a one-time install that automates the same protocol through hooks and alone sends every state, blocked and note included; it is returned only when tool is set (claude-code, codex or antigravity), otherwise hook_install_note says so; run_wrapper puts lastping run before a launched command (cron job, CI step, script) and reports start, success, fail and cancel. Also returned: failure_inbox_how_to, expectations_how_to (declare_run_expectations), tracing_how_to (OpenTelemetry, detailed by get_trace_setup), otel_env_lines, export lines whose key placeholder stands for the person's tracing key, and docs_url. An exporter that cannot set headers can POST to <ping_url>/v1/traces, which needs no Authorization header.

Requires the read scope or higher. Gets one run's full timeline: every recorded event (start, step, log, success/fail/cancel, incident_opened) in time order, its declared assertions with pass/fail/not_evaluated verdicts against the terminal ping body, the terminal output excerpt, CI metadata when present, and its OTLP spans (spans[], parents before children, siblings by start time) when traced. A current Claude Code hook's run holds its turns' traces; a prompt after an interrupt or a blocked turn continues the open run. Also carries project, receiving_spans, in_hook_session, is_test, failure_cause and upstream_error (as in list_runs). For the detail behind a run (id + rid) found elsewhere. outcome: succeeded, failed, cancelled, blocked, running or unfinished (no end ping, no incident, not blocked, older than max_runtime_s or 24h; never pages). Caps: 200 events (events_truncated; the terminal event is kept), 2,000 spans (spans_truncated, spans_dropped). A span's gen_ai block (system, model, tokens, cost_usd) is present only on GenAI calls. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Gets a monitor's run history from pings, CI and agent runs alike (no filters; trace-only runs are in list_runs). Each run has rid, kind, received_at, steps (seq, name, at; matched on rid), title and incident_detail. CI runs add failing_stage, actor, commit_sha, run_url, branch, duration_s and outcome. A ping with neither ci_meta nor a rid is excluded. duration_ms is LastPing's own /start-to-success timing, separate from CI's duration_s. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Why traces, metrics or logs sent to one monitor did or did not arrive: the newest 20 ingest attempts (kept 7 days), last_accepted_at, and the newest traced run's summary. For checking a test span, or telemetry sent with nothing showing. Each attempt: outcome (accepted; refused: something to fix; dropped: routine, answered 202), reason, span_count, bytes, protocol, user_agent, signal. Rejections: unsupported_media_type (not http/protobuf, e.g. gRPC to the HTTP URL), body_too_large (over 1 MB), too_many_spans or too_many_records (over 500 per batch), unknown_monitor (lastping.monitor_id missing or outside the project), expired_key (a new tracing key comes from the Connect page), wrong_scope (not a tracing key), wrong_project, monitor_mismatch, over_budget/over_log_budget (daily, resets 00:00 UTC), rate_limited, busy, malformed. Refused after a 202: future_start (sender clock), too_many_series. Dropped: unknown_event, unknown_metric, cumulative_temporality, invalid_point. No row: gRPC to the gRPC port, or a missing or wrong key. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Returns the steps that make a tool send OpenTelemetry traces to one monitor: what to write and where, how to verify it, and the tool's limits. For setting up tracing, observability or telemetry. prompt is the full set-up procedure for the named tool, credential handling included. Each block has files, a key_line (the terminal line that stores the tracing key the person creates on the monitor's Connect page) and, for Claude Code and Codex, a reviewable bang_script run with bang_command.

Requires the read scope or higher. Lists the project's agents: id, slug, name, status, monitor_count, last_seen. status is rolled up live from the agent's monitors, worst first: down, blocked (a run needs a human), late, running, up, pending (no report yet) or idle (no monitors, or all paused or in maintenance). usage_24h: model tokens and cost over 24 hours (null with no model call); top_dependencies: its five heaviest outgoing dependencies. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the admin scope or higher. Lists the project's API keys without plaintext values. Each has its non-secret prefix, scope (read, write or admin), created_by_key_id (the key that minted it, absent for a dashboard key; revoking a key revokes every key below it), last_used_at and last_used_surface (mcp, terraform or api; both absent when never used). last_used_surface comes from the caller-controlled User-Agent header: a hint, not proof of identity.

Requires the read scope or higher. Lists recent alert deliveries across the project's monitors: whether an alert fired, failed or was suppressed, and to which destination. Each row is one (incident event, destination) outcome: pending (attempt in flight), delivered, dead (per-channel attempt ceiling reached) or suppressed (the destination's rate cap dropped it). Covers 30 days and returns only the newest page; the dashboard's delivery log has the archive. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Everything the project's agents call, across every agent, most calls first: each dependency with the same figures as get_agent_dependencies plus agents (which agents call it, and how often). Answers questions such as which agents call postgres or use a model. At most 50 rows; more counts the rest. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Lists the project's notification destinations (channels) of every kind (webhook, telegram, discord, slack, ntfy, pushover, msteams, googlechat, email). Their ids are what set_route takes.

Requires the read scope or higher. Trace sources that sent spans but match no registered agent: id, source_name (the OpenTelemetry service.name), first and last seen, span_count, and suggested_agent_id when a registered agent's slug or name now matches. For traces that arrive while no agent shows them; adopt_discovered_agent counts a source under an agent. At most 200, most recently seen first. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Lists a monitor's recent incidents, newest first; an open incident has closed_at=null. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Lists every monitor in the project with id, name, slug, status and ping_url. tag narrows the list to monitors carrying one tag.

Requires the read scope or higher. Returns an agent's failure inbox: every open incident on the monitors it owns, newest first, with context no single failure body carries. For learning, at a run's start, what broke meanwhile. failure_signature.occurrences: how often this exact failure has been seen (first_seen, last_seen, fingerprint), which separates retry from escalate. failed_step: the last step reported (for 'stalled', where the live run is stuck). exit_code: 137 (usually OOM) and 1 differ. duration_vs_normal: e.g. '8.2x the typical run (41m vs 5m), from 30 archived days', plus run_ms, typical_ms, ratio, days_sampled (0: the norm is the median of 5+ recent runs). cause: 'fail' (the job reported an error), 'silence' (it never reported; usually the scheduler or host) or 'upstream' (consecutive model-provider API errors; detail names the last). Also body_excerpt, run_id, ci.run_url. A missing field is no evidence, not none, normal or a clean exit; exit_code 0 marks a success that failed its expectations. add_incident_note takes an entry's incident_id. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Lists runs across every monitor in the project, newest started first, including runs that exist only as OpenTelemetry traces (traced: true). Each run carries check_id, check_name, rid, title, outcome, started_at, ended_at, duration_ms, step_count, exit_code, its incident, and when traced span_count, tokens, cost_usd and cost_source (client, estimated or mixed); plus agent_id, agent_name, source_name, multi_trace, project (the Claude Code session's folder), receiving_spans (spans in the last 2 minutes), in_hook_session, is_test (excluded from counts). failure_cause is 'upstream' when the run ended on the model provider's API error (pages after 3 in a row, or failure_threshold when higher; upstream_error names it) or 'assertion' when a success failed an output assertion or declared expectation. outcome: succeeded, failed, cancelled, blocked, running or unfinished (started, never ended within max_runtime_s or 24h; never pages). Filters combine and also narrow counts (per-outcome totals). Pages with next_cursor. untrusted_fields names the data fields LastPing did not write (job output, exporter data or user-supplied names).

Requires the read scope or higher. Lists the project's status pages: id, slug, title, the monitors on each, visibility, and the public URL of any public page. A status page shows a monitor's health to people outside the project, such as customers or another team. update_status_page's check_ids replaces a page's monitor set, and this result holds the current set.

Requires the write scope or higher. Pauses a monitor (paused=true): it still receives pings but raises no alert.

Requires the admin scope or higher. Replaces an API key's secret: a new key, with a new id and the same name, scope and (for a tracing key) monitor. The old key stops working immediately wherever it is held, including when it is the calling key. No expiry stays no expiry; an expiring key gets a fresh 90 days, capped at the calling key's expiry. Keys the old key created keep working (no cascade). Refused (403, with max_scope) for a key scoped above the caller's. The plaintext key appears only in this result.

Requires the write scope or higher. Registers an autonomous agent in the project's agent registry and returns its id, slug and wire-up steps. One agent stands for one autonomous worker, which can own many monitors: create_monitor and update_monitor attach a monitor through agent_id (the id or the slug). An agent_id that does not exist is an error (400 UNKNOWN_AGENT), never an implicit create. The slug is derived from the name, so registering the same name again is refused (409) rather than creating a second agent.

Requires the write scope or higher. Resumes a paused monitor (paused=false); alerting resumes from the next missed ping.

Requires the admin scope or higher. Permanently revokes an API key and every key it created, recursively; all of them stop authenticating immediately. This cannot be undone; list_api_keys' created_by_key_id shows which keys a revoke reaches.

Requires the write scope or higher. Sets or clears one alert message template on a monitor; every other template is kept (read-modify-write). The template is validated for allowed variables before saving, and an empty template resets that entry to the built-in default. Event types: 'down', 'recovery', 'fail', 'every-run', 'success', 'started', 'blocked', 'note'. The template argument lists the variables.

Requires the write scope or higher. Routes a monitor's alerts for one event type to a set of destinations (channels). Replaces the whole destination set for that event type: destinations not listed stop receiving it, including ones someone else configured. get_monitor's routes field holds the current set, so adding a destination means sending the existing ids plus the new one. An empty channel_ids removes all routing for the event. Destinations have to be verified and enabled (an email one confirmed). list_destinations returns the ids.

Requires the write scope or higher. Sets or clears a maintenance window on a monitor. During it, deadline incidents (a missed or never-started run, an overrun, a stall, a blocked run outliving blocked_timeout_s) and, on an HTTP monitor, failing probes (any fail there counts as a probe's) are recorded but open no incident; a site still failing afterwards opens one on its next failure. Still notified: the job's own fail ping, the page a blocked ping queues, a runaway ping rate, and routed note, started, success and every-run events. Takes exactly one of duration, until or clear=true.

Requires the write scope or higher. Sends something through a destination now. By default it delivers a synthetic 'LastPing test alert', which confirms a new destination's credentials. With resend_verification=true it instead re-sends an unverified email destination's confirmation link, for when the first one never arrived or expired; an unverified email cannot join a route, and a test alert does not change that.

Requires the write scope or higher. Updates an agent's name, description or slug by UUID with merge-patch semantics: supplied fields change, omitted ones keep their value. Renaming never changes the slug. A new slug has to be unique in the project; saved links, Terraform references and trace sources (service.name) naming the old slug then stop matching the agent unless they equal its name (case-insensitive), past traced runs included, and a slug equal to a source another agent receives takes that source's traces, past runs included. Attached monitors stay attached.

Requires the write scope or higher. Updates a notification destination's name and/or config in place; only the supplied fields change. The kind cannot change (delete and recreate instead). A new email address resets verification and sends a new confirmation email.

Requires the write scope or higher. Updates a monitor by UUID with merge-patch semantics: supplied fields change, omitted fields keep their stored value. tags, assertions (conditions a successful run's ping body has to satisfy, which catch a job that exits 0 having done nothing) and guards (ceilings on a number the job reports, which catch a looping agent) each replace the whole set; get_monitor returns the current sets. slug and ci_provider are immutable; only the ci_workflow and ci_branch filters change here.

Requires the write scope or higher. Updates a status page's title, slug, visibility or monitor set; arguments not supplied keep their value (the tool reads the page and merges, so an omitted check_ids never blanks it). A supplied check_ids replaces the whole monitor set; list_status_pages returns the current ids. A new slug changes the public URL and breaks links already shared.