The gateway / Console API
Console API
The endpoints that exist for a screen of the console and for no worker.
The console is one client of the gateway; every screen is an endpoint of Gateway API or of the pages beside it. This page is the endpoints that exist for a screen and no worker:
| endpoint | scope | what |
|---|---|---|
GET /v1/usage?after=&limit= |
usage |
the org's metered rows after the cursor, {rows, totals, next}; one row per call.summary and call.score folded, the cursor the store's position so a billing consumer resumes losslessly |
GET /v1/insights?day=&days=&agent= |
calls |
whole UTC days of the key's environment and scope: the window of days (1, 7 or 30; 1 by default) ending on day (today by default), and with agent that agent's calls alone. {day, days, timezone, conversations: {now, before}, resolved_rate, median_e2e_s, spend_usd, channels: {phone, web, whatsapp}, judged, passed, escalated, mean_length_s, endings: [{reason, count}], series: [{day, phone, web, whatsapp, spend_usd, judged, passed}], sessions_total, live, agents: [{slug, calls, score, spend: {llm_usd, stt_usd, tts_usd, phone_usd, platform_usd, minutes, per_minute_usd}}], stages: [{stage, vendor, model, turns, median_s, p95_s, confidence}], budget: {limit_usd, spent_usd_month}}. before is the calls of the window of the same length right before; judged the calls a judge settled and passed those no judge broke; escalated the calls a person took part in; mean_length_s over the calls that ended; endings how they ended, the most first; series one row per day of the window, a day with no call included. Any other days is a 400. The budget is the org's across both environments, and a call's cost counts in the month (UTC) its summary was written, the month of the window's last day; stages is each stage of the window's turns by the vendor and model its report names (see docs/wire/metrics.md, "Per day"); an agent's spend is what its calls of the window cost by stage, off their summaries' rows (the platform's own compute rows under platform_usd), and per minute of the calls that ended: what to optimise first |
GET /v1/insights/drift?agent=&before=&after= |
calls |
what moved for one agent of the key's environment and scope between two sides, each a day (2026-09-29) or a version of its settings (v3); by default today against the day before, in UTC, and a side that is a version needs the other named. The answer is {agent, world, before, after, judges, stages, versions}: each side as asked with the versions its calls ran ({day, version, versions}); each judge with {held, judged, pass_rate} on each side, moved (the pass rate after minus before) and criteria_changed (the question it asked is not the same text on both sides, so the move may be the judge's, not the agent's); each stage by vendor and model with {turns, median_s, p95_s, confidence} on each side and median_moved_s, p95_moved_s; the most moved first. versions is every version either side ran or that was set between the first day's start and the last day's end, each {version, holder, author, note, set_at, ran_before, ran_after}: what changed and who changed it. Read from what each call's seal counted (docs/wire/metrics.md, "Per day"), so a call still going is not in it |
GET /v1/limits |
any key | each quota of the key's environment as {limit, used} (minutes, messages, llm_tokens, concurrent_calls, agents, seats, numbers, evals, simulations), lends, billing_url and world; used reads what admission reads, so the page and a refusal agree |
GET · PUT /v1/org/judging |
calls · usage |
whether the org's calls are judged at hang-up, the model they are judged on (model, a ModelConfig; null: the platform's), and the platform's ceiling per call in dollars |
GET · PUT /v1/agents/{slug}/widget |
talk · pipeline |
how the widget presents the agent per environment: {title, tagline, greeting, accent, autostart, theme}; a null field is the widget's own default, an accent is a CSS colour |
GET /v1/agents/{slug}/pipeline and the hold melody |
pipeline |
Pipeline |
GET · PUT /v1/agents/{slug}/settings, /v1/agents/{slug}/lexicon |
pipeline · words |
Settings |
GET /v1/providers, /v1/provider-keys, /v1/voices, /v1/voices/sample |
providers · pipeline |
Provider keys |
POST /v1/agents/{slug}/dev/{family}/{verb} |
the family's | Dev verbs |
/v1/agents/{slug}/personas, /v1/evals/* |
evals |
Evals |
/v1/org/judges, /v1/agents/{slug}/judges |
evals |
the org's judges and an agent's own: Evals |
/v1/knowledge, /v1/contacts/{contact}/memory, /v1/agents/{slug}/memory |
knowledge · memory |
How retrieval works |
| the threads, the line, the numbers, the accounts | WhatsApp · Numbers · People |
Money is US dollars everywhere: a call's cost, the judge's, a budget.