Console page directory
For: People looking up a console page.
Find the page below for its tabs, actions and permission limits.
Overview
Four figures over the period: jobs, success rate, p95 duration (with its median) and cost, compared with the period before (insights). Activity opens first with the period's jobs and filters for status and agent; Trends shows the time series; Cluster shows servers and edges. The address keeps the selected tab.
Approvals and nodes that need attention appear above the figures; each number links to its page. For a tenant's admins, a card of the secrets past or near their expiry.
On a new setup, Get started: the steps your role can take, each with its command and the page where it is done, ticked when the server says it is done (a cluster admin joins a node, adds a model key, installs a verified plugin, runs a job and sets a budget; a tenant's admin adds a model key, installs a plugin and runs a job, joins a node while the tenant has none, and as an owner invites people; an operator runs a job; a reader has no steps).
Collapse steps reduces the checklist to progress and the next step; View steps opens it again. Hide hides it in this browser, and Show setup steps, at the top of the Overview while it is hidden, or in ⌘K, brings it back.
While the setup steps are expanded, Try a first job: a card for each verified starter (net-diag, host-health, the PR reviewer) with what it needs, whose Start runs it and opens the job, or, for an admin, Install and start opens the install with its verified listing and starts it after the install, or, when the install warned, after the admin read the warnings and clicked Start; both go away with the checklist (first jobs)
Jobs
The jobs that started in the period (the last 7 days unless you pick another), newest first, a page at a time, filtered by status and agent; one word in the search finds a job by its ID, and more words or agent:, tag: or group: search the job summaries of the period, which reach back past Temporal's retention, by agent, status, tag and text in their errors.
The page shown is read again every five seconds and stays the page shown; click a row to peek at a job without leaving the list. A job's page has a tab for its steps, a waterfall with findings, where a step you click opens its details under its own row, as Temporal's event history does, and a second click folds them; its live stream while it runs, child jobs on other nodes included; its answer; and its cost per branch.
A running job says why it waits, and what waits for a person shows above its tabs with the buttons to decide it. The header has one action, Cancel while the job runs or Run again once it failed (compare); the rest is under More: Debug opens it in the studio's debugger, Export downloads it as JSON, Compare puts it beside another job, and its history in Temporal's own UI, which admins reach at /temporal/.
Its ID is under the title with a copy button; operators tag a job; beside the tabs are the job's facts and its child jobs
Chats
Your chats as a full page (/chats), the list beside the open chat; the same chats as the chat drawer beside every page
Approvals
What waits for a person, a tab each: Waiting, the steps of the tenant's jobs that wait now (no period: they all wait in the present), searched by what, agent or job on the server and a page at a time, which you open in the dialog to approve or deny, with a comment; Budget requests, the budget requests that wait for the tenant's or a team's admins; and Changes for admins, the changes asked for in the period (the last 30 days unless you pick another), filtered by state: waiting, applied, rejected or failed.
There is no history page of decided approvals: the audit log has each decision
Agents
The agents of the tenant's plugins and where they can run, searched by name and description and filtered by plugin, and a Run button that starts a job with an input and a budget; an agent's dialog links to its jobs and its deep dive: runs, failure rate, latency and cost over time, versions, tools, models, nodes and errors
Tools and skills
The tenant's tools with their functions and the nodes that run them, and on a tab of their own its skills, each searched and filtered by plugin; a tool's name opens its deep dive. Each function's badge says what the tool policy does with its calls and links to Policy, where the policy is set
Plugins
Installed plugins with their version, commit, nodes and quality gate, searched by name; a plugin's dialog has its source, tools, agents, the gate's report and its status on each node. Install from a Git URL or an uploaded bundle, update (one from a plugin index to the newest version listed there), remove; MCP Registry finds a server of the MCP Registry and installs it (MCP Registry)
Routines
The tenant's routines with their triggers, state and last fires, searched and filtered by state; an editor checked as you type, a dry run, Run now, pause and resume, and each routine's fires of a period (the last 7 days unless you pick another), a page at a time
Studio
The tenant's workspace: write plugins with lint at each line as you type, forms that keep comments, revisions and their diffs, the plugin's graph, installs with the gate's verdict; a test bench that runs an agent or a flow live or with recorded answers; a debugger that steps through any job (the studio)
Findings
Jobs whose steps look wrong: loops, retries, idle time, waits for a node, budgets, workflow failures, in the period (the last 24 hours unless you pick another), filtered by finding (any warning when none is chosen) and agent; a job's dialog has all its findings, each with its fix where there is one, its cost and a link to its steps
Errors
Failures of jobs and their steps grouped by what they say, in the period (the last 7 days unless you pick another), most jobs first, filtered by agent, tool and node, with counts, first and last seen, the agents, nodes and versions they hit, and their jobs (errors)
Compare
One page at /compare, a tab for each comparison: Jobs, two jobs side by side with their steps aligned (?a=<job>&b=<job>); Versions, two versions of an agent with their numbers, errors, evals and changed files (?tab=versions&agent=<agent>); and Drafts and revisions, an agent's draft or revision tried against what runs (?tab=drafts) (compare). It measures quality; what a setting saves is Measured savings, a page of its own
Policy
Under Govern, the rules every tool call meets, each part as visible as it was before it moved here: Rules, the tool policy with each rule's own decider and deadline, its lint and what its shadow rules would have done (every member reads it; admins change it); Egress, the sandbox egress list of each tenant; Replays, a backtest of a change on recent work; Statistics, how approvals are decided; and under More, Four eyes, the AI label and Approval defaults, the tenant's default approval deadline and escalation ladder
Kill switches
Under Govern, for those who read governance: Active, what is stopped across the fleet, with a form to stop an agent, a tool or the tenant and to release it for those who may change governance; Receipts, the signed receipt of each stop and release, for those who read the audit log
Privacy
Under Govern, at /governance: retention and the legal hold for those who read retention (admins, auditors, privacy officers), a person's data to find and erase for those who may, and for every member Labels (the tenant's label list and its label policies, in one tab), Erasure (the erasure policy and keep-for periods) and Works council (works-council mode)
Inventory
Every agent, flow and tool with its owner, purpose, risk class, models and their regions, and the data it handles, searched and filtered by kind, plugin, risk class and owner, with counts of the agents that have no owner person, have been unused for 90 days or have no risk class; admins edit them in the dialog, and Export downloads the whole inventory as CSV or JSON (inventory)
Audit log
Under Govern, for the tenant's admins and its auditors: who changed what, what the platform did by itself, and what runs did, in the period (the last 7 days unless you pick another), newest first, filtered by target, action and, for cluster admins, the tenant shown or every tenant and the cluster; a search by who started a job tree takes a reason, which is itself recorded; an entry's dialog has its detail and its place in the hash chain.
Verify checks the chains, Export downloads one as JSON lines (audit trail)
Nodes
Every node with its state, health, runtime, labels and last heartbeat, filtered by status (a cluster admin with several tenants may choose every tenant's) and searched by ID, name, label or host; a node's dialog adds its machine, certificate and health penalties. Accept, reject, block, unblock and delete. Add nodes creates a join token and shows the two commands to run on the machine, ready to copy.
Upgrade rolls a runtime out to every node, a canary first, and a panel above the nodes follows each step to its verdict, with Roll back while it runs. For the tenant's admins, a tab of the join tokens (a cluster admin's are every tenant's), each revoked there
Queues
Every task queue with its backlog, its oldest task, the nodes polling it and those that may, and why one is stuck, searched and filtered by state and kind (queues)
Cluster
Under Fleet: Temporal backends, the backends the cluster has and where each tenant's workflows run, with a move to another backend for cluster admins; Regions, the cluster's regions and each tenant's, for cluster admins
Cost
Four summary metrics, then Usage with one Group by selector for agents, models/services, nodes or jobs. People and teams shows ownership and your budget; Efficiency shows spend anomalies, model speed and prompt caching; Outside spend shows costs imported by plugins. The period, tab and grouping stay in the URL
Budgets
Under Spend, what the tenant may spend and who is told when: Tenant budget, each tenant's monthly budget beside this month's spend (its admins set it); Quotas, the quotas of teams, people, keys and services with this month's spending, which admins set and members ask more of; Alerts, the spend alerts at the shares people choose and the daily spend-anomaly finding, read by operators and changed by admins with a reason. Budget requests in the header opens them on Approvals, where they are decided
Models and providers
The provider records the gateway routes by, how each signs its calls (a key, an OAuth2 token or SigV4) and whether another tenant uses the same key, the tenant's model aliases and its data rules; admins change the tenant's own
Access
Under Admin, the one place for people. For cluster admins and owners, People: every account, searched by name, with its cluster role, its roles in each tenant, how it signs in, its second step and when it was last seen; a person's dialog sets their roles in the tenants you own, signs them out, resets their second step or their password, and links the account to an identity provider or unlinks it; Sign-in methods, the identity providers and where a second step is required; and Roles.
For every member, Teams: the tenant's teams with their members, searched by name, each team's quota shown as it is with a link to Budgets (admins make and remove teams and members). Under More, Sessions, every signed-in browser, for cluster admins, and API keys for the tenant's admins: the API keys (shown once, searched by name, and filtered to those of a removed account), admin keys, the apps of the token exchange and the tenant's CORS origins
Credentials
Under Admin, for the tenant's admins, the one place for secrets and connections. Secrets: each secret set, tested with its providers first, rolled back within 24 hours, given an expiry and deleted, with its version, age, setter and expiry, searched by name, and what reads it (plugins, model providers, credential bindings and where they inject it, the connection that saved it); the secrets something reads that are not set yet, each with Set it; and a shared provider key (secrets).
Bindings lists every credential binding, searched by name. Connections, for those who manage connections: every connection a plugin declares, searched by name, whose dialog has its params and steps, Connect, what it signs in as, when a call last got its token and the plugins whose granted uses act through it (connections).
Refusals: the last 24 hours'. Credential changes that wait show above the tabs; never a value (oz0 secret usage)
Settings
Settings holds the tenant-wide switches no other page owns:
- Tenants: each tenant and its namespace; cluster admins create them.
- The tenant's encryption status, visible to every member, with Rotate payload key for admins.
- Notification rules and your own address.
- Ask, the chat that finds the right agent.
Under More, you can manage the MCP clients you connected with OAuth. Owners also see everyone's clients and control which clients may connect. Your own account settings cover your password, your second step and where you are signed in.
Operators also see Plugin settings. Set a setting here when no plugin reads it. A setting one plugin reads appears with its current value and a link to that plugin's Configure tab, where it is set.
Connect an MCP client is the page an MCP client sends you to: it names the client and where its answer goes, and you approve or deny it