The console
The screens your team works in: the floor, calls, quality, testing and settings.
The console is where your team watches calls, reads them, tests agents and changes their settings, in a
browser. It is one page for both production and the sandbox: production at https://cloud.pinecall.io/, the sandbox at
https://cloud.pinecall.io/sandbox/, on one sign-in.
How it works#
The sidebar has four parts, and your name at its foot. What you look at comes first and reads the same for every agent or one. What builds an agent appears when one is in view. The Harness — simulations, personas, cases, judges and monitors — is how an agent is put to the test and watched, for every agent or one. The workspace, at the foot, is the org's.
Viewing, at the top of the sidebar, picks whose calls are on screen: every agent, or one. Changing it keeps the screen you are on. Production · Sandbox opens the same screen on the other side.
Open it#
$ pinecall console
console https://cloud.pinecall.io/sandbox/?login=lc_9f2 (opens within five minutes, once)pinecall console opens the sandbox's console in your browser, already signed in. --prod opens
production's, if your org lets you act there. pinecall console <agent> opens that agent's screens. Your
key never travels: the link carries a one-use code that is good for five minutes.
The screens#
| Screen | What it shows | Where |
|---|---|---|
| Overview | live calls with Listen and Take the line; calls, resolved without a person, held by the judges, median answer and spend over 24 h, 7 d or 30 d; Needs a look | both |
| Calls | every conversation as threads or as a table; filters for live, wants a person, did not hold | both |
| Quality | the share of judged calls where every judge held, a day at a time, what broke, by agent | both |
| Agents | every agent held, its channels and numbers, the processes holding it; Apps: the agents Pinecall runs for you (pinecall deploy), their releases, logs, stop and start; Secrets: the values those apps start with |
both |
| Knowledge | Docs: every base of the world, its size and which agents search it; Memory: every fact any agent's calls taught — the same tabs an agent has, with every agent's | both |
| Numbers, Team, Usage | the org's numbers, people and roles, and what it consumed — minutes, tokens, evals and simulations among it | production |
| Privacy | Rules: how long a call is kept, when and how often a number is rung, where consent is asked, what a call says first; Consent: consents and the do-not-call list; Export & erase: the world exported, a contact erased, the trail of every erasure; Access log: who read what | both |
| Observability | Day by day: latency end to end and by stage (median and p95), how calls ended, each judge's held rate, tool failures, spend — every agent's or the one in view | both |
| Settings | what is set once: Tokens, Providers, Telemetry (a copy of every spoken call's trace to a tool of yours), Alerts (a webhook of the org's own for every alert), and in the sandbox Phone testing (which number reaches your sandbox copy) | both |
Notifications is yours, not the org's: click your name at the foot of the sidebar. It is what the bell tells you — a call asking for a person, every call, a monitor that fires — on this browser and your phone.
Harness, its own section of the sidebar: each row is one screen, the same with every agent in view or one — only what it lists narrows:
| Row | What it shows |
|---|---|
| Simulations | the form — agent, persona, line — and every simulation already run down the left; the one open beside them, heard live and scored at hang-up |
| Personas | Pinecall's library first, each with a switch, then the agent's own; the one open beside them, with what it expects, how it acts on the line, and its runs; Generate… has the judge model write callers from the agent, kept one by one |
| Cases | the groups — what broke the agent, one row a judge, each diagnosed into a cause and a fix: Apply the setting, Open in your assistant, Make a persona, decided whole — then every case, real or simulated, down the left; the one open beside them |
| Judges | every judge a call meets at hang-up: Pinecall's library with a switch on each (the org's, or the agent's in view, which wins), then the org's own and the agent's own, written and edited here; Try asks one of the agent's last calls and writes nothing; the Judging card turns judging on or off and names the judge model, Pinecall's or one of yours |
| Monitors | the numbers watched over a window, the line each must not cross, and the last day each fired |
Every filter is in the URL, so a view is a link you can paste to a colleague.
An agent's screens#
With one agent in view, Build adds what makes it:
| Row | Tabs |
|---|---|
| Playground | Chat with the agent by text or voice; Dev chat in the sandbox, for the class running in your terminal |
| Test | Overview: the six stations of the harness, each with its number, and the latest runs; Goldens and their runs |
| Knowledge | Docs: the bases it reads; Memory: what its calls taught |
| Configure | General: voice, models — the judge model among them, in its Judge tab — language, with its Memory and Knowledge sections (the policy, and what it knows by heart); Pipeline, Lexicon, Widget |
Goldens' Run all, a case's Run it and Dev chat work with the pinecall start holding the agent, in its
directory: the class runs there. With no such process the card says so and the rest of the screen loads. Simulations
call the agent as anybody calls it, so whatever holds it answers — deployed, or pinecall start.
Who sees what#
A screen is drawn only for a key that opens it, so nobody meets a refusal on a click:
| Scope | Opens |
|---|---|
calls |
Agents, Calls, an agent's Overview |
evals |
Quality, Test, Goldens, Simulations, Personas, Cases, Judges, Monitors |
talk |
Playground, Dev chat, Widget |
words |
Configure, Lexicon |
knowledge, memory |
Docs, Memory |
numbers, team, usage, pipeline, providers |
the screen of that name; team also opens Data & privacy; providers also Telemetry and Alerts |
supervise |
the desk on a live call: listen, whisper, take the line |
Orgs, people and roles says which role holds which scopes.
Turn on notifications from the bell in the top bar. A call where the agent asked for a person is the one thing a supervisor must not miss, and production tells you about it by default once the browser allows it.