Agents Overview

Agents automate repetitive legal tasks in August. Define the task in natural language, and August plans execution steps. Run agents on demand or on a schedule, and track progress through activity logs. The objective is consistency and time savings on recurring work.

This video demonstrates how to build and run agents for repetitive legal tasks.

How agents are created

You create an agent by describing it in chat. On the Agents page of a project, click Create New to open a chat that drafts the agent for you, then click Save when the preview looks right. See Creating Agents from Assistant for the full flow.

After saving, you can change the agent two ways from its page:

Edit reopens the chat so you can describe changes.
Edit manually (in the menu next to Edit) lets you change steps, inputs, and outputs directly on the canvas. See Build and edit agents on the canvas.

You need edit access to the project to create agents, and edit access to an agent to change it. People you share an agent with as viewers can run it but not edit it.

What an agent is made of

Every agent has inputs, a sequence of steps, and one or more outputs.

Inputs: documents, folders, playbooks, tabular reviews, other agents, or questions the person running it answers. Saved defaults are applied to every run.
Steps: each step is one of five types: Analyze (review and understand information), Create Artifact (produce a document or other deliverable), Take action (use a tool or connected app), Question (ask a person for input), or Branch (choose what happens next based on a rule). See Agent Control Steps for Question and Branch steps.
Outputs: the deliverable the run must produce, in Text, File, or Table format. See Agent output contracts and branch-specific outputs.

Running an agent

Open the agent from the Agents page and click Run. If the agent asks for inputs, a Run agent window opens first; otherwise the run starts immediately. To run it inside a chat instead, open the menu next to Run and choose Run in Assistant. See Run an agent for details.

You can also start runs from the August public API. See Trigger an agent run from the API.

Creating Agents with Natural Language

Describe what you want accomplished in everyday language:

"Every Monday, review new documents in the ABC Matter folder and generate a summary of key developments."
"When new contracts are uploaded to the NDA folder, extract term, renewal, and termination provisions into a table."
"Daily: Check for new emails from opposing counsel and flag any mentioning deadlines."

Refine the request if outputs are too broad or too narrow. Use specific scope statements to improve repeatability.

Running agents automatically

Triggers run an agent without anyone clicking Run. Open the agent and click Triggers to set them up:

Schedule: run hourly, daily, on weekdays, weekly, or monthly.
Upload: run when files are uploaded to an August folder you choose, including its subfolders.

Permissions, batching, and scheduling rules are covered in Agent Triggers.

Monitoring Agent Activity

The agent page lists past runs under Run History, labeled Manual, Scheduled, or Triggered. Open a run to follow it step by step on the run graph:

See which steps completed and in what order
Identify where adjustments are needed
Review outputs at checkpoints, especially for high-impact tasks

A run that pauses on a Question step shows as Waiting until someone replies. See Resume an agent run waiting on human input.

Common Agent Patterns for Legal Teams

Agents follow pass-based patterns: orientation, issue spotting, and summary output.

First-pass Diligence Summaries

Upload a data room, run an agent to extract key terms across all agreements, and generate an executive summary. Review the structured output, then drill into specific documents for deeper analysis.

Issue-Based Document Triage

Set an agent to scan incoming documents for specific provisions (assignment clauses, change of control triggers, termination rights). Flag any that require attorney attention.

Internal Update Generation

Schedule a weekly agent to summarize developments in a matter folder and produce a status update for the deal team or client.

When to Use Agents

Use agents when:

Tasks are repeatable and structured enough for consistent execution
Inputs follow a predictable pattern (same document types, same extraction goals)
You want to reduce manual effort on high-volume, lower-judgment work

Escalate judgment-heavy or novel issues to attorney review. Agents handle the repetitive work so you can focus on decisions that require legal expertise.

Best Practices

Start narrow: Begin with a focused scope and expand once the agent proves reliable.
Validate early runs: Review outputs carefully before putting an agent into production.
Keep prompts concise: Specific, well-defined instructions produce more consistent results.
Use checkpoints: For multi-step agents, build in review points to catch issues early.