Tour of the console

Every screen at tools.armoriq.ai, what question each one answers, and where to go when you want a specific thing.

The console at tools.armoriq.ai is where you read everything your assistants did. It does not make decisions itself. It shows what was recorded.

The sidebar carries six screens: Home, Analytics, AI Chat, Sessions, Intent Plans, and Policies. Several more screens exist and are reached by links from those pages, or by typing the address. Press ⌘K anywhere to search policies, agents, and pages.

Home

The first screen after sign-in, and the one to check daily. Until your first run arrives it shows the setup timeline instead. After that it becomes Workspace overview, with five counters: intent plans and tokens over the last 24 hours, estimated spend and actions prevented over 30 days, and how many things are awaiting you.

Below that, Today's briefing lists actions held for a human decision, and Recent runs shows the latest activity with a verdict on each.

The counters describe recorded activity. They do not measure how much went unreported, and the page says so rather than implying full coverage.

Sessions

A session is one stretch of work with your assistant. This screen lists them with totals for the workspace you are in, sorted by token spend by default. Use the sort toggle for most recent instead. Filter by harness (Claude Code, Codex CLI, Gemini), search the loaded rows, and sort by tokens or by recency. Each row expands to show cache reads and writes, how many model rows the session has, and when it was first seen. The per-model breakdown itself is on the session detail. View activity on any row opens the session detail.

Session detail

The screen that answers "what did it actually do?". It has two halves, deliberately kept apart.

Recorded session usage is imported transcript usage: processed tokens, input including cache, output, cache reads, and a row per model.

Runtime evidence is the tool and policy record. It holds the MCP call ledger, oldest call first, filterable to failed calls, retried calls, or calls with a policy link. Select any call to open an evidence drawer with its execution outcome, timing, policy result, and plan link.

Above the ledger, a three-step evidence chain summarises the session: call evidence, retry and timing, then policy result. Underneath are four tables: Tools with call counts, errors and average duration, MCP servers, Policies with allow, deny, hold and ask counts, and Models with tokens in and out.

Policy links are reported honestly as one of three states: linked on this span, linked by call id, or no policy link. A missing link means evidence was not reported. It is not proof the agent was unprotected.

Analytics

Tokens, estimated cost, and latency, sliced the ways you actually ask for.

Pick a range (24 hours, 7, 30, 90 days, or all time), then filter by harness, by repository, and by whether Armor was on, meaning a policy or tool was evaluated in that session. The daily chart toggles between cost and tokens.

The tiles cover processed tokens, cached input, uncached input, output, and cache savings. A latency card gives p50 and p95 turn wall time per harness, including any time the machine was asleep mid-turn, which is why occasional outliers look large. Below that are breakdowns by model, by session, and an Agent session activity table you can filter to one agent or one session UUID.

Costs are estimates from configured model rates, not bills. Usage with no configured rate is excluded from cost and the page tells you how many tokens that affected.

Policies and Policy Studio

Two different things, kept on separate screens because they enforce in different places.

Policies lists server-side policies that apply to agents and MCP servers, with status tabs, a decision filter, priority sorting, and card, list or timeline views.

Policy Studio is where you author the rules your local plugin enforces: per-tool allow, block or hold, a folder lock, and the default for anything unmatched. This is the screen most people want.

Tools

Which tools are configured, which were actually observed, and which were evaluated by ArmorIQ. Filter by product, or by evidence: observed in 24 hours, guarded in 24 hours, errors in 24 hours, or unused for 30 days. A Connection evidence card at the top reports whether an active device key exists and whether a correlated evaluated session was found. The wording is careful on purpose: an active key is not proof of a live connection, and configuration is inventory evidence, not live reachability.

Runs

Executed agent actions across products, newest first, with a verdict on each: Running, OK, Blocked, Held, Failed, or Unknown. Filter by product, agent, or verdict, search across action, tool, agent and session, toggle extra columns, and Export copies the table to your clipboard as CSV.

Intent Plans

Every agent turn becomes a signed intent plan. This screen tracks how closely each run followed its plan and where it drifted, with tiles for plan count, on-plan rate, held, and blocked. Each row expands to show steps executed, risk level, model, tokens, cost, and the stated intent.

Inbox

Workspace notifications and actions held for a human decision, split into today and earlier.

You cannot approve or dismiss a held action here yet. The console tells you what is waiting; resolving it happens in the agent workflow that raised it.

AI Chat

Ask about your recorded usage and policy activity in plain language, such as "how many tokens have I used in 4 days?" or "which policies were most blocked in 30 days?". Answers stream back with evidence cards that name their source, the window, and when the data was retrieved. Chat history is not saved. Each conversation lives in the browser tab until you leave it.

Other screens

Machines at /users: which machines are spending tokens, ranked, with cost estimates.
Graph at /graph: a map from workspace to product to session to tools, MCP servers, and policy decisions.
Agents at /agents: agents registered in the workspace, with a register dialog.
MCP Servers at /mcp-servers: register, scan, and monitor MCP servers and their tools.
Prompts at /prompts: reusable prompt templates with {{variable}} substitution. These are saved only in your browser, are not synced to your team, and are instructions rather than enforced policy.

Settings

Opens as a dialog over whatever you were looking at. Profile and Appearance cover your account: name and email come from the provider you signed in with, and theme and density apply instantly on this device. General renames the workspace, which admins only can do. API Keys manages keys for the SDK and CLI. Billing holds your plan, the monthly or yearly choice, Stripe checkout, the customer portal and your invoices, with the invoice list limited to admins. Install repeats the install commands so you can add another machine without leaving the console.

On this page