Agent Discovery
Your organization is already running AI agents — in Microsoft Copilot Studio and Azure AI Foundry, in Google's Vertex AI and Gemini Enterprise, in Amazon Bedrock AgentCore, in MSP platforms like Hatz AI, and in the assistants, builders and automation tools teams adopt on their own. Most of them are invisible to security and IT: no inventory, no activity trail, no policy.
Agent Discovery is the neutral, cross-platform inventory of those agents. Connect a platform once — read-only — and Waxell:
- Discovers every agent, assistant and AI workflow on it into one fleet inventory.
- Observes their activity — who ran what, when, with which model — from each platform's own audit and telemetry, with no changes to the agents themselves.
- Shows how governed each agent is, from dark (discovered, no signal) to enforced (Waxell has proof it gated the agent), and gives you the levers each platform exposes.
Credentials are read-only and encrypted at rest. Where a platform supports it, Waxell connects with no stored secret at all — an admin-consent click, a keyless role grant, or a cross-account role gated by an External ID.
Supported platforms
Twenty-one platforms connect today, grouped the way they appear in the connector picker. The connector catalog lists what each one discovers, what you need to connect it, and what Waxell can do beyond discovery.
| Category | Platforms | Guide |
|---|---|---|
| MSP platforms | Hatz AI | Connect Hatz AI |
| Cloud agent platforms | Microsoft 365 Copilot & Copilot Studio, Azure AI Foundry, AWS Bedrock AgentCore, Gemini Enterprise Agent Platform (Vertex AI) | Microsoft · Google Cloud · AWS |
| AI assistants & workspaces | Glean, Atlassian Rovo, Salesforce Agentforce, ServiceNow AI Agents, Box AI, Dust | SaaS connectors |
| Agent builders | Devin, Mistral Agents, LangGraph Platform, Botpress | SaaS connectors |
| Workflow, iPaaS & RPA | n8n, UiPath, Workato | SaaS connectors |
| Code & repositories | GitHub, GitLab (repo scan) | SaaS connectors |
| Observability signals | Datadog LLM Observability | SaaS connectors |
More platforms — Zapier, Make, Retool, Microsoft Power Automate, Automation Anywhere, Blue Prism and others — are listed in the picker under Coming soon. Choose one and select Request access to tell us you need it.
The Agent Discovery workspace
Agent Discovery has six tabs, plus a page for every connected platform and every discovered agent.
Overview
The fleet at a glance: how many platforms are connected, how many agents were found, and the governance coverage of the whole fleet — the share of agents in each posture band. Platform cards are grouped by posture and open that platform's page. The Most active agents list surfaces the agents doing the most work.
Connections
Everything you've connected, and where you add more.
- Each connected platform shows its state, how many agents it found, and when it last synced. Select a row to open the platform page; select Settings to change credentials or disconnect.
- If a sync fails you see a plain-language reason — for example "Couldn't reach Azure — the connection was dropped. Retry the sync." — with the technical detail available under Details.
- Add connection opens the connector picker.
The connector picker
The picker is a searchable catalog of every platform, organised by category, with Recommended platforms first and MSP platforms at the top. Each tile tells you, before you click:
- What it finds — for example "Copilot Studio and M365 Copilot agents, their connectors and conversations".
- What connecting takes — API key · ~2 min, 1-click admin consent, Keyless (grant a role), IAM role (CloudFormation).
- What Waxell can do there — Activity (pulls audit/telemetry), Policy (can write governance config), In-path (can sit in the request path). A solid badge means live today; an outlined badge means the platform's API supports it.
Search matches names, vendors, product names and categories — "copilot studio", "bedrock", "rpa", "msp" all work. Use the arrow keys to move between tiles, Enter to open one, and Esc to close.
Choosing a platform opens its connect panel: numbered setup steps, a link to the vendor's documentation, the fields to fill in (optional ones are marked), and the button to connect. Platforms with a key check — such as Hatz AI — verify your key with the vendor before saving it, so a wrong or expired key is caught immediately.
Registry
Every discovered agent in one table — name, platform, type, owner, posture and last seen. Filter by platform, posture band, kind or status, sort any column, and search by name or owner. Filters are part of the page address, so a filtered view can be bookmarked or shared.
Each row shows a next step matched to the agent's posture: Instrument a dark agent, Scope a policy for an observed one, or Promote to enforcing for a governed one.
Incidents
Policy findings on discovered agents — which policy fired, the action taken (block or warn), the tool or connector involved, the agent and the acting user — each linked to the run it came from. Filter by platform or policy.
Activity
The raw activity Waxell pulls from each platform's audit and telemetry APIs — sign-ins, agent changes, invocations, tool calls. Filter by platform or event type. When an event belongs to a discovered agent, it links to that agent. Activity refreshes automatically every 10 minutes.
Settings
Fleet-wide behaviour, including auto-disable on a block violation for agents on platforms that support it.
Platform pages
Every platform has its own page — open it from the Connections tab, the Overview, or any agent. A platform page shows the connection's state and last sync, with Sync now, Pull activity and Build runs where the platform supports them, and five tabs:
| Tab | What it shows |
|---|---|
| Overview | Headline numbers and the platform's own breakdown — customer tenants and model mix for Hatz AI, environments and connectors in use for Microsoft, AWS accounts for AgentCore, surfaces and models for Vertex, models and agent kinds for Foundry — plus the busiest agents and latest activity. |
| Agents | The platform's agents with columns specific to it (for example customer tenant, 30-day runs, users and models for Hatz AI; environment for Copilot Studio; AWS account for AgentCore). |
| Activity | That platform's activity feed, filterable by event type. |
| Controls | What Waxell can do on this platform — observe, govern and in-path capabilities — and its enforcement targets, such as Foundry model deployments, the Vertex Model Armor floor setting, or AgentCore gateways. |
| Health | The connection's history — created, verified, first sync, last sync — its authentication mode, and its capabilities. |
A platform you haven't connected yet shows what it would discover and how to connect it.
Agent pages
Select any agent to see everything Waxell knows about it:
- Posture — its band, why (the signal sources Waxell matched and when each was last seen), and what moves it up.
- Platform details — the fields that matter for that platform: tools and knowledge sources, models, customer tenants, environment, framework, runtime identity and so on. Every raw field is available under All metadata.
- Connectors & tools it uses, recent executions, activity attributed to it, and the policies scoped to it.
- Controls where the platform supports them — for Microsoft Copilot Studio agents, the kill switch and the Waxell policy hook.
Governance posture
Every agent sits in exactly one band. The same definitions are used everywhere in Agent Discovery.
| Band | Meaning |
|---|---|
| Enforced | Waxell has proof it gated this agent. |
| Governed | A Waxell policy or hook is scoped to it, not yet proven enforcing. |
| Observed | Waxell sees its activity, but nothing governs it yet. |
| Dark | Discovered, with no live signal. |
| Unknown | A signal source was unavailable, so Waxell couldn't tell. |
Governed coverage, wherever it's shown, means Enforced + Governed. Sample agents are never counted.
How sync works
- Connect. Add the platform and authorize it with the action that platform uses. Waxell shows you the exact values to use.
- First sync. Waxell discovers the fleet. For most platforms this runs as soon as you connect. For platforms that onboard many customers at once, such as Hatz AI, the key is verified immediately and the first sync is scheduled — the connection shows First sync scheduled and results typically appear within one business day.
- Ongoing. Inventory and activity refresh automatically every 10 minutes. Agents that disappear from a platform are kept as history and marked inactive, so your inventory never silently loses an agent.
For MSPs
- One key, every customer. MSP platforms like Hatz AI connect with a single MSP key that covers every customer tenant you manage. Each agent is tagged with its customer, and the platform page breaks usage down per customer.
- Or one workspace per customer. Scope a connection to a single customer tenant when each customer has its own Waxell workspace.
- Org-wide clouds. The cloud connections scale to a whole estate in one step — Microsoft tenant-wide consent, a Google org-level role grant, or an AWS Organizations StackSet.
Revoking access
Disconnect a platform from its Settings panel. Waxell stops syncing immediately, and you can also revoke the grant on the platform side — delete the key, remove the consent, or remove the role. Discovered inventory is removed with the connection.