How it works
Here's a closer look at what happens at each step, from the moment a conflict is found to the moment it's resolved.
The process
01
Detect
Conflict detection runs automatically the moment new information is added, comparing it against everything you've already documented. It's not a separate step you have to remember to run. It's part of ingestion itself.
02
Triage
Every pair is scored into a severity tier before it ever reaches a person, so an urgent contradiction doesn't get lost in a pile of minor ones. Near-duplicates and compatible phrasings are filtered out before they ever surface.
03
Decide
You're shown where each claim came from and how reliable that source has been before you choose. Arthos never makes the call for you. It makes sure you're not making it blind.
04
Resolve
Your decision becomes the answer going forward, and stays linked to its source. Future conflicts that resemble it can point back to how this one was resolved, so your judgment compounds instead of resetting every time.
Integration
You decide how you use Arthos. If you'd rather just see conflicts surface and make the call without touching code, we can build that interface with you. If your team wants to integrate directly, everything is also available through the CLI or the API. Either way, it's built to fit how you work, not the other way around.
If you'd rather just see conflicts surface: here's an example of what that could look like.
Step 1 of 2: Review the conflict
Flagged contradiction
The Riverside office is closing permanently at the end of the quarter. Source: "Riverside Office Closure Memo," Facilities, Aug 12, 2026
The Riverside office lease was just renewed through 2029. Source: "Riverside Lease Renewal Amendment," Legal, Jun 30, 2026
2 similar mentions elsewhere in your documents were reviewed and found to be saying the same thing as one another. This is the only one that's a genuine contradiction, not just different wording.
Step 2 of 2: Choose what's true
Both can't be true. What happens if you pick each one?
An illustrative example, not a screenshot of a fixed Arthos screen. We'll work with you to integrate it into your tooling and your team's own interface.
If your team wants to integrate directly: every action is available as both a CLI command and a REST API call.
What you get back from GET /api/conflicts/<id>:
{
"id": "c_48a1",
"status": "open",
"severity": "YELLOW",
"claims": [
{
"text": "Remote employees are reimbursed for home internet up to $50/month.",
"source": "Employee Handbook, v3",
"reliability_rating": "high"
},
{
"text": "Remote work stipends were discontinued as of Q1.",
"source": "HR Policy Update, March 2026",
"reliability_rating": "high"
}
],
"resolution_options": ["keep_canon", "make_new_canon", "counter_point", "rectify"]
}
Under the hood
Deciding which version of the truth is correct is only as good as what you can see behind it. That's what this is for.
Down to the line, not just the document.
Each fact links back to the exact passage it came from, not just which file it was in.
You decide how much of that detail you see.
A quick citation is enough day-to-day for some teams; others want the exact line and timestamp surfaced every time. The precision is always there either way.
Sources carry a reliability rating that adjusts over time.
That's part of what informs your decision when two sources disagree. We work directly with you to tune what "reliable" means for your organization, since that's not the same for every customer.
Questions
Is this a knowledge graph, or a search tool?
Not really. A graph shows you how things relate to each other. Search helps you find something you already know you're looking for. Arthos does something different: it actively looks for places your own knowledge disagrees with itself, tells you what's at stake if that conflict stays unresolved, and helps you settle on one source of truth.
Do I use Arthos as a one-time audit or consulting engagement?
No. Think of it more like a copilot, or a partner, as you develop your policy or narrative. It stays with you the whole way, so a new contradiction doesn't sit unnoticed for months until someone happens to run a review.
Do you provide the interface our team uses day to day?
Either. We can provide the interface, or integrate directly into yours through the API or CLI, whichever fits how your team already works.
How do you know something is a real contradiction and not just two ways of saying the same thing?
Severity is tuned to what your organization actually considers urgent, not a fixed global rule. Arthos learns your organization's priorities over time, so what surfaces reflects your own judgment, not a generic default.
Who owns the data we share with you?
Your data is always yours. Specifics on handling, retention, and security are things we work out directly with each partner, including a data processing agreement where needed.
We're early, and this works best as a two-way build. Let's talk about what you're working with.
Let's partner