DrozLegal / Product / AI Agents

An AI workforce, built for Canadian law.

DrozLegal’s agents are named, observable, retry-bounded jobs that do specific work for specific lawyers — not a chat box you negotiate with. They run on a real queue, report their token spend, and finish their work with an audit trail you can read.

7 production agents Async job runner Token-cost metering per job APScheduler triggers
Named workflows

The seven that run today.

Each agent is a real named workflow with a defined scope. You initiate it; it finishes the work; the Briefing surfaces the result.

a01

bulk_intake

Ingest a batch of inbound matters — from a portal, an email folder, or a CSV — triage them, run conflict checks against the CRM, and open matters with engagement letters drafted.

Triage · Conflict · Open matter
a02

ar_dunning

Nightly via APScheduler. Reviews aging A/R, drafts gentle reminders at 30/60/90 days in the firm’s voice, and posts them to the Briefing for sign-off before send.

Scheduled · Nightly
a03

wills_outreach

Scans wills and POAs aging 5+ years and drafts re-engagement letters to clients. Turns dormant files into renewed mandates.

Revenue generation
a04

minute_book_review

Quarterly minute-book audits per corporate client. Flags missing resolutions, missed annual returns and out-of-date registers.

Compliance · Quarterly
a05

contract_review

Risk-flags inbound contracts against your firm’s playbook, proposes redlines, and explains each change in the margin in your associate’s voice.

Playbook-aware
a06

defence_drafting

Generates Statements of Defence end-to-end from the Statement of Claim — with Rule 4.01 formatting, paragraph-by-paragraph admit/deny logic, and backsheet attached.

Ontario civil litigation
a07

motion_preparation

Assembles the motion record — Notice of Motion, supporting Affidavit, Factum with authorities tabbed, Costs Outline at close — ready for CaseLines upload.

Motion-record automation
+8

Supporting agents

course_generation · discovery_review · form_recommendation · intake triage · redline_explanation · smart_routing · suggest_changes · time_entry

In production support roles
Inside the engine

The console where the work actually runs.

Every agent is a job. Every job has a lifecycle. Every step is observable.

app.drozlegal.ca / agents / console
Active jobs
14
2 in retry
Completed today
38
Across 6 agents
Tokens spent
2.4M
$18.40 cost
Awaiting sign-off
9
Routed to Briefing
Running · queue depth 4
running
contract_review · Cascadia Hydro — 47 pp. master services agreement
00:42 in
running
minute_book_review · Belanger Holdings — Q2 audit, 3 missing resolutions
01:08 in
running
defence_drafting · Tremblay v. Northway — Statement of Defence, paragraph 27 / 41
02:14 in
queued
wills_outreach · 14 dormant files queued for re-engagement letters
scheduled 11:00
queued
ar_dunning · nightly run, 22 invoices in 30/60/90 buckets
tonight 21:00
Completed · needs sign-off
done
motion_preparation · Singh v. Loblaw — record assembled, 6 authorities tabbed
09:42
done
bulk_intake · 7 new matters opened, 1 conflict flagged
10:11
review
contract_review · Ardent Series B — 4 high-risk clauses redlined
10:48
review
defence_drafting · McAllister Trust — awaiting partner sign-off
11:30
Engine infrastructure

What we built underneath, so the agents don’t fail quietly.

  • Full job lifecycle — queued, running, completed, failed — surfaced in the agent console.
  • Token-cost metering per job, per matter, per lawyer.
  • Parallel subagent architecture: compliance / financial / regulatory / risk reviewers each contribute to the final output.
  • APScheduler time-based triggers for nightly and weekly workflows.
  • Retry bounding so a failed agent doesn’t burn tokens infinitely.
  • Timeout protection on every step in a long-running workflow.
  • Async LLM job runner separating compute from request lifecycle.
  • Structlog structured logging and Sentry error tracking.
  • Redis rate-limiting on outbound model calls.
  • License kill-switch middleware: a workspace can be paused mid-job without losing state.
Try the agents

See an agent finish your next file. On one matter. In a week.

Pick the agent that hurts the most — contract review, motion prep, minute books, A/R — and watch it run on a real file before your firm commits to anything.

SOC 2 Type II PIPEDA & Law 25 Data in ca-central-1