Justinian is for
Agent platform
Government has two AI problems and they need different agents. The legal problem is judgment: read the statute, find the controlling authority, draft the rule, defend the decision. The operational problem is throughput: process the permits, registrations, and renewals at the volume a state actually generates without the workflow breaking. Justinian is built for the first. Trajan is built for the second. They share the corpus, the control plane, the audit trail, and the security posture, so an agency that procures one can deploy the other without running a second procurement.
Vulcan agent stack
Rulemaking + permit backlog response
Advise
01Justinian reads the corpus, surfaces controlling authority, flags conflicts and deadline exposure, and produces the legal record behind the decision.
Execute
02Trajan intakes the citizen request, identifies the right agency workflow, gathers missing documents, and prepares the package for the human reviewer.
Approve
03Filings, agency decisions, outbound messages, and record updates stop for human approval before they are taken.
Audit
04Every action logs the source passages, tools, files, and human authorization. The record holds up to FOIA, IG review, and opposing counsel.
Justinian
Legal and policy advisor
Reads the corpus, surfaces controlling authority, drafts rulemaking support, compares jurisdictions, and gives counsel the legal record behind the decision.
Trajan
Government AI concierge
Turns a citizen request into the right agency workflow: intake, requirements, missing documents, status, and a package ready for reviewer action.
Security layer
Approvals and audit
Consequential actions stop for human approval. Every answer and workflow event ties back to the source passages, tools, files, and users involved.
The work is grounded before it is generated.
The agent starts from the relevant corpus, matter files, and user-approved context. It does not ask a model to improvise from memory.
The output is delivered with the record attached.
Sources, tool calls, files, approvals, and final work product are connected, so a lawyer, reviewer, or agency leader can reconstruct what happened.
Coverage
Most legal AI tools query someone else's database (Westlaw, Lexis, a public scraper) through an API, with rate limits, gaps, and a layer of someone else's editorial choices between the agent and the law. Vulcan doesn't. We built the corpus. We maintain it. We update it. Our agents read it directly.
6,245,376,303 records. 141 corpus types. Federal, state, and municipal: primary law, courts, public finance, health, environment, education, labor, transportation, agriculture, disclosure filings, and the operational data that govern public programs. Everything indexed, parsed, and citation-linked, so the path from question to source to answer is real and reviewable.
6,245,376,303 records across 141 corpus types. See the full corpus page →
Why Vulcan
Every Justinian answer ties to a primary source the agent actually read. Every Trajan decision ties to a published rule, a completed workflow, and a human approver. Every action across both products leaves a record an inspector general could reconstruct, a FOIA office could produce, and opposing counsel could cross-examine. Built for the offices that have to defend the work, in public, in court, and in front of the legislature.
Public-sector agents need more than clever answers. They need accountable action. Who authorized the task. What the agent could access. What it actually did. Which human approved the result. Vulcan is designed around those four questions because every government office reading this page already has to answer them, and the software has to make answering them easier, not harder.
Agent identity
Each agent run is tied to the authenticated user, workspace, permissions, and task that authorized it.
Least-privilege tools
Tool access is scoped to the matter, agency environment, deployment, and action class needed for the job.
Approval gates
High-impact actions can stop for human review before a filing, outbound message, decision, or record update.
Structured audit trails
Source passages, retrieved files, tool calls, generated artifacts, and approvals remain attached to the output.
Private data boundary
Each deployment runs in the customer tenancy. Integrations are scoped, reviewable, and revocable. Privileged matter, citizen records, pre-decisional materials, and protected categories never leave the boundary.
Enterprise access controls
SAML SSO, data lifecycle controls, logging, and reviewable security posture fit procurement and counsel review.
Company
Vulcan is building agents that can safely do serious work: read private records, operate across systems, produce artifacts, and leave a defensible trail. The next site should make that ambition obvious from the first scroll.
AI for government, answered
What AI agents does Vulcan build for government?
Vulcan builds two agents for government: Justinian, a legal AI agent for statutes, regulations, policy, fiscal, and legislative work, and Trajan, a citizen-facing government AI concierge for permits, registrations, license renewals, forms, and government services.
What government work can Justinian handle?
Justinian supports permitting, licensing, registration, procurement, fiscal analysis, legislative drafting, regulatory review, and agency legal work. It reads the governing statutes, administrative code, and agency records together and returns cited, verifiable work product.
Is Vulcan secure enough for government use?
Vulcan is FedRAMP Ready and runs in AWS GovCloud (US), architected to the FedRAMP Rev. 5 control baseline and independently assessed by a 3PAO. Agent runs are authenticated, tools follow least privilege, humans approve sensitive actions, and every step lands in an audit trail.
What is Trajan?
Trajan is a government AI concierge for citizens: a one-stop shop for permits, registrations, license renewals, forms, laws, and regulations. It is the model behind efficiency.texas.gov.
Demo request
Bring a real question. A live rulemaking, a backlog you can't process, a statute your counsel is wrestling with, a permit type your agency is buried in. We arrive with the first analysis run, the first workflow mapped, or the first citations pulled, whichever helps most.
Read the coverage behind our $10.9M seed round.
Trusted outlets across government, legal, and technology press covered Vulcan's seed announcement. Explore the full set of stories below.