How Fusial's AI works

Fusial uses AI to make contract review faster, not to replace the reviewer. Every accept, reject, counter, and export still requires the user's explicit decision — the AI's role is to surface the relevant facts and draft suggestions for you to approve or discard.

What the AI does

• Extracts counterparty, classification, summary, and key terms when you run AI analysis.
• Identifies section structure for navigation.
• Runs redline passes against your playbook and surfaces issues with severity.
• Suggests responses (accept / reject / counter) to incoming counterparty edits.
• Drafts replacement clause text on request.

The three AI pipelines

Every AI feature in Fusial is one of three named pipelines:

Extraction

Runs when you click Run AI analysis on the contract page. Required before the redline and response pipelines can run.

• Reads the parsed paragraphs and runs.
• Produces structured output: counterparty, classification, summary, key terms, and section outline.
• Each extracted value is anchored to the paragraph it came from, so you can click through to the source.

Redline

Runs when you click Run AI redline, calibrated against your organization's playbook.

• Walks the contract clause by clause.
• For each clause, checks whether it aligns with your playbook context (jurisdiction, compliance frameworks, governing-law preference, dispute resolution preference, company description, custom notes).
• Returns one issue per finding: a category, the offending clause, severity, rationale, and a suggested rewrite.

Response

Runs when you click Review their edits, available whenever a contract has pending incoming edits.

• Looks at the counterparty's tracked change in context.
• Recommends an action — accept, reject, or counter — with a short rationale.
• For counters, Fusial drafts replacement text you can edit before sending.

Model tiers

Fusial routes AI calls through two internal tiers:

• Fast — used for high-volume, low-stakes calls like per-edit response suggestions and incremental extractions.
• Full — used for higher-stakes reasoning like a full redline pass or extraction on a long document.

Pipelines request a tier rather than a specific model, so the underlying model can be upgraded over time without changing how Fusial behaves on your contracts.

How AI output stays grounded in your document

Every AI output in Fusial is tied back to the source contract:

• Stable paragraph anchors. Every paragraph gets a unique ID injected at upload. The same ID survives re-parses and is the unit of reference for every AI output.
• Citations on every output. Each extracted term, each redline issue, and each response suggestion references the paragraph(s) it was drawn from.
• Click-through to source. From any AI output in the UI, you can jump straight to the highlighted source paragraph in the document view.
• Rewrites are presented as drafts. Counter-text and redline rewrites are shown as suggestions you apply, so the document only changes when you decide.

Auditing AI activity

Every pipeline run is recorded in the workspace audit log. Open Settings → Audit log to see, for each AI pass:

• Who triggered it.
• Which pipeline ran (extraction / redline / response).
• When it started and finished.
• Whether it completed or failed.

Filter by date, action, or user to narrow down events, search by user or contract title to jump to a specific run, and expand any row to see the raw payload — including edit IDs and decision metadata.

Your controls

• Members and above can trigger any AI pipeline. Viewers can read AI output but can't initiate a pass.
• The redline pipeline can be re-run at any time from the editor header (Re-run AI redline).
• The response pipeline can be re-run whenever pending incoming edits remain on the contract.
• To stop sending data for a specific contract, delete the contract — the source .docx, parsed body, and all AI outputs are removed together.

Related