A1 · Strategy & Research

Research you can trace

Continuously scans market size, trends, user demand signals, and regulatory shifts relevant to the product vertical. Every claim keeps the passage it came from, and every fetch is one a policy allowed.

Start free Open the workspace

What it does

Four steps, in this order, every run. Nothing is fetched before a plan exists, and nothing is written up before its evidence is checked.

Step 1Plan the researchA brief becomes a plan: the questions to answer, the sources worth reading, and what counts as an answer. You approve the plan before anything is fetched.
Step 2Fetch the allowed sourcesOnly hosts in the project's own source registry are reachable — deny by default, exact host match. Anything resolving to a non-public address is refused outright.
Step 3Validate the evidenceEach claim is tied to the passage it came from, and contradictions between sources are surfaced rather than averaged away.
Step 4Analyse and write upMarket size, trends, demand signals and regulatory shifts, assembled into a report with its coverage and uncertainty stated, not implied.

Inside the workspace

The agent has a workspace of its own, not a single results pane — the brief, the plan, the evidence behind each claim, and the reports, each at its own address.

Nine tabs per research projectBrief, Plan, Run, Evidence, Reports, Market, Monitors, Tasks and History — each a sub-route, so a reload or a shared link lands exactly where you left off.
Source registryThe hosts this project's research may reach, and nothing else. It is the fetch allowlist, not a bookmark list.
Templates and librariesReusable brief templates and curated libraries of sources, claims and reports.
Five analytics dashboardsQuality, usage and cost, operations, source usage, and a read-only research API with per-key spend.
Shareable reportsA read-only link for one report, readable with no XOOZLY account. The token is the whole credential, and every refusal reads the same.

How it is governed

A research agent reads the open web on your behalf. These are the limits it runs inside — all of them enforced in the platform, none of them a prompt instruction.

Routed through the MCP gatewayEvery tool call passes policy check, then audit log, then dispatch. There is no second path to a tool, and no domain can bypass it.
Server-side request forgery, refusedA host is resolved once and fetched at the address that was checked. Redirects are refused unless a bounded number is asked for, and each hop is checked like a first request.
Polite by constructionOne request per host per second by default, a host's own rate from the registry where it has one, and robots.txt Crawl-delay honoured. A host that keeps failing is paused.
Spending stops where you set itA project's cost cap is a hard limit checked before a run starts, with span-level tracking and an automatic pause on breach — not an end-of-month reconciliation.

What it costs

The agent is included in every tier — there is no separate licence for it. You bring your own Anthropic, OpenAI, Gemini or OpenRouter key and pay the model directly, with no markup from us.

Start free Compare the tiers

What it does

Four steps, in this order, every run. Nothing is fetched before a plan exists, and nothing is written up before its evidence is checked.

Step 1Plan the researchA brief becomes a plan: the questions to answer, the sources worth reading, and what counts as an answer. You approve the plan before anything is fetched.

Step 2Fetch the allowed sourcesOnly hosts in the project's own source registry are reachable — deny by default, exact host match. Anything resolving to a non-public address is refused outright.

Step 3Validate the evidenceEach claim is tied to the passage it came from, and contradictions between sources are surfaced rather than averaged away.

Step 4Analyse and write upMarket size, trends, demand signals and regulatory shifts, assembled into a report with its coverage and uncertainty stated, not implied.

Inside the workspace

The agent has a workspace of its own, not a single results pane — the brief, the plan, the evidence behind each claim, and the reports, each at its own address.

Nine tabs per research projectBrief, Plan, Run, Evidence, Reports, Market, Monitors, Tasks and History — each a sub-route, so a reload or a shared link lands exactly where you left off.

Source registryThe hosts this project's research may reach, and nothing else. It is the fetch allowlist, not a bookmark list.

Templates and librariesReusable brief templates and curated libraries of sources, claims and reports.

Five analytics dashboardsQuality, usage and cost, operations, source usage, and a read-only research API with per-key spend.

Shareable reportsA read-only link for one report, readable with no XOOZLY account. The token is the whole credential, and every refusal reads the same.

How it is governed

A research agent reads the open web on your behalf. These are the limits it runs inside — all of them enforced in the platform, none of them a prompt instruction.

Routed through the MCP gatewayEvery tool call passes policy check, then audit log, then dispatch. There is no second path to a tool, and no domain can bypass it.

Server-side request forgery, refusedA host is resolved once and fetched at the address that was checked. Redirects are refused unless a bounded number is asked for, and each hop is checked like a first request.

Polite by constructionOne request per host per second by default, a host's own rate from the registry where it has one, and robots.txt Crawl-delay honoured. A host that keeps failing is paused.

Spending stops where you set itA project's cost cap is a hard limit checked before a run starts, with span-level tracking and an automatic pause on breach — not an end-of-month reconciliation.