AI Agent Overview
Introduction to the MarkDocket AI agent — what it does, how it fits into the platform, and the key concepts users need to know before using agent chat.
The MarkDocket AI agent is a conversational assistant built into the platform that can research, analyze, and act on your IP portfolio data in real time. Instead of navigating individual screens, you can describe what you need in plain language and the agent will invoke the right tools, interpret the results, and stream a response back to you.
What the agent can do
The agent has access to the same catalog of tools available through the CLI and MCP integrations. During a conversation it can:
- Search trademark and patent databases, including USPTO filings and prior art
- Retrieve and summarize portfolio items, deadlines, watches, and clearance reports
- Kick off research workflows such as brand clearance or due-diligence analysis
- Query your vault files and attach documents to a turn for multimodal analysis
- Look up goods-and-services classifications and suggest refinements
- Perform web searches for competitive or prior-art context
Not every tool available in Automations is available in interactive chat. A small set of agent-only tools — including web search and goods classification — are exclusive to the agent loop and do not appear in the CLI or MCP tool listings.
How the agent fits into the platform
The agent sits behind the MarkDocket API, which acts as the gateway between your browser session and the AI service. Your conversation is routed through the API to the agent, and every action the agent takes on your behalf calls back through the same API layer that handles all other platform operations. This means:
- Your data permissions apply. The agent can only access resources your account is authorized to see. It cannot escalate beyond your own permissions.
- All actions are audited. Every agent turn — including the tools it called and the files it accessed — is recorded in your account's session history.
- Billing is pre-authorized. Before the agent begins processing your message, the platform reserves the necessary credits. If your account does not have sufficient credits, the turn is denied before any work begins and you are prompted to top up.
Agent turns
Each time you send a message in agent chat, the platform creates an agent turn. A turn moves through several stages:
| Stage | What happens |
|---|---|
| Authorizing | Billing is checked and a credit reservation is made |
| Authorized | Files you attached are validated and accepted |
| Running | The agent loop executes, calling tools and streaming results |
| Completed | The turn is finalized and the credit reservation is settled |
If something goes wrong — such as a network interruption after you press send — the platform is designed so you can retry safely. Each turn is identified by a stable key generated at the moment you initiate it, so retrying the same send does not create duplicate charges or duplicate file uploads.
If your account runs out of credits mid-turn, you will see a billing notification in the chat interface. The turn stops cleanly and your session history is preserved so you can resume after adding credits.
File attachments
You can attach files to an agent turn directly from your Vault or by uploading from your device. Attached files are scoped to the turn they were sent with. The agent can read and reason over their contents — for example, reviewing a specimen image or analyzing a draft specification — as part of its response.
Tool catalog and what the agent can act on
The agent's available tools are derived from the same canonical registry used by the CLI (@markdocket/cli) and the hosted MCP endpoint. This means that any tool you can call from the CLI is also available to the agent in chat, plus a few additional tools that are agent-session-specific.
The agent does not have direct access to external data sources. All lookups — USPTO records, brand-intel results, vault files — go through the MarkDocket platform, which means results are subject to the same data freshness and rate considerations as the rest of the platform.
Bring Your Own Key (BYOK) and the agent
If you have configured a BYOK AI provider key in your account settings, be aware that BYOK keys apply only to Agent nodes inside Automation Runs. Interactive agent chat always uses the platform's AI service. This is enforced at the platform level and cannot be overridden per-session.
Automations vs. agent chat
Agent chat and Automations both use the agent's tool catalog, but they serve different purposes:
- Agent chat is interactive and synchronous. You send a message, the agent responds, and you can follow up immediately. It is best for exploration, one-off queries, and tasks where you want to review and guide each step.
- Automations are pre-configured, durable workflows that run on a schedule or trigger. Each node in an automation graph runs as a checkpointed step, so long-running workflows can survive restarts and resume from where they left off. Agent nodes inside automations can also use BYOK keys.
For tasks you run repeatedly — such as weekly watch-report summaries or automated clearance checks for new filings — an Automation is more appropriate than repeated agent chat sessions.
Privacy and data handling
Messages and tool results in agent chat are stored in your account's session history. The agent does not retain information between separate chat sessions — each session starts fresh. Files you attach to a turn are stored in your Vault and remain accessible after the turn completes.
Session history — You can review past agent sessions and the tools called during each turn from the Agent section of your dashboard. This history is visible only to members of your account with appropriate access.
Next steps
- Start a chat — Open the Agent panel from your dashboard to begin your first session.
- Explore the tool catalog — See the CLI tool reference for the full list of tools the agent can call.
- Set up Automations — Learn how to turn recurring agent tasks into scheduled workflows in the Automations overview.
- Configure MCP — If you use an external AI assistant, connect it to your MarkDocket data via the hosted MCP endpoint.
Hosted MCP Endpoint
Reference page for the MarkDocket hosted MCP endpoint: how to connect over OAuth 2.1, how access tokens are scoped, and which tools are available compared to the stdio server.
Agent Turn Lifecycle
Documents how an agent turn progresses through its full lifecycle: submission phases, billing pre-authorization, file upload, streaming, and completion or cancellation.