Risk comes from interactions
A provision that looks harmless alone can create exposure when combined with a definition, exception, basket, or release term.
Behavioral infrastructure for AI systems
AI without rules creates chaos.
DAC turns agreements into rules AI can follow.
Agreements define what people, software, and AI systems may, must, and must not do. DAC makes those rules usable by AI while keeping every result tied to its source and human authority.
ClauseFlow · Institutional credit
Exposure and opportunity rarely sit in one clause.
Most AI tools identify which clauses exist. ClauseFlow traces how the clauses work together, shows the path visually, and links every finding to the governing language.
Every material finding remains linked to the governing source and subject to expert review.
Standard AI can miss the chain
Clause-by-clause review can miss how one provision changes the effect of another.
ClauseFlow makes the chain reviewable
Follow the cited path through the agreement and inspect the language behind each finding.
Why DAC exists
Agreements already define the rules.
They set authority, conditions, boundaries, consequences, and remedies.
As AI systems join human networks, they need a usable form of those rules. DAC preserves the source and human authority behind each decision.
May the drill engage at these coordinates?
Conditional · §4.2
The drill may engage because the coordinates fall within the authorized extraction zone and torque, thermal, and power conditions remain inside mission limits.
Agreement
system
system
Run-to-run consistency
Same agreement. Same question. AI alone may take a different path each time.
An AI model rereads and reasons over the agreement on each run. ClauseFlow follows the same saved provision path, keeping every finding tied to the same cited language for review.
Standalone AIRUN 01 / 03
Yes, the agreement appears to permit it.
ClauseFlowRUN 01 / 03
Conditional: only if capacity remains. No separate consent condition appears in the cited pathway.
What ClauseFlow adds
More coverage. Faster first review. Knowledge the team can reuse.
ClauseFlow expands the terms and interactions a team can review, brings clause-grounded evidence forward sooner, and preserves findings across the portfolio.
Structural elements per agreement
Track vulnerabilities, protections, blockers, calculations, and connected pathways across each agreement.Broader structural coverage
Compared with the seven to nine headline issues commonly prioritized in manual review.First-pass visibility
Bring clause-grounded analysis forward before the transaction, amendment, trade, or legal queue moves.Institutional memory
Create a queryable record of negotiated terms, protections, structural pathways, and precedent across the agreements under coverage.How DAC works
Turn a credit agreement into a connected, reviewable analysis.
ClauseFlow connects the provisions that shape an answer and lets the reviewer check it against the agreement.
The same source-linked structure supports questions, calculations, updates, and approved workflows without rebuilding the agreement for every prompt.
Governing source
Agreement + version
Rights + obligations
Conditions + exceptions
Representations + warranties
Events + remedies
Resolved stateCovenants + restrictions
Improve existing AIReuse the workAPI / MCP
Every result links back to the agreement, clause path, and version.
Why credit first
Corporate credit depends on agreement interpretation.
Investment decisions across origination, underwriting, portfolio management, trading, and restructuring with support from risk and legal depend on how the agreement is read.
ClauseFlow is built for institutional corporate credit, where connected provisions can change the outcome of an investment, amendment, or restructuring.
CLO and leveraged-loan managers
Query protections, pathways, concentrations, amendments, and precedent across large loan portfolios.
Private-credit and direct-lending platforms
Test negotiated protections, sponsor requests, amendment language, structural capacity, and accepted precedent.
Credit-focused and multi-strategy asset managers
Connect downside underwriting, relative value, distressed analysis, special situations, and legal diligence.
Banks and dealer platforms
Support leveraged finance, loan capital markets, trading, restructuring, legal, and risk workflows.
ClauseFlow
Follow the path from clause to exposure.
ClauseFlow connects each provision in the chain, shows the interaction visually, and links the result back to the agreement.
ClauseFlowcited agreement analysis
Ready for a questionProject Helix Credit Agreement · 412 pages
Project Helix Credit Agreement
The Borrower may designate any Restricted Subsidiary as an Unrestricted Subsidiary if, immediately after giving effect, the designation is permitted as an Investment under Section 6.04.
Investments may be made in an aggregate amount not to exceed the greater of $500,000,000 and 26.0% of Adjusted EBITDA, subject to the conditions stated herein.
Upon any Subsidiary becoming an Unrestricted Subsidiary, the Liens on its property shall be automatically released without further action by any Secured Party.
Agreement
Credit agreement
document set
Conditional right
Designation permitted?
§5.13
Capacity test
Capacity remains?
§6.04(q)
Automatic effect
Lien release automatic
§9.09(d)
Not in cited path
If consent were required
counterfactual protection
Potential exposure
Potential collateral release
Ask a question…
Product preview uses a fictional agreement and an illustrative scenario. Every answer remains subject to expert review.
Formula Composer
See how one definition changes the result.
Definitions, baskets, ratios, and covenants depend on one another. Change one input to see its effect on capacity, leverage compliance, headroom, and any connected pathway.
Every supported number remains linked to the clause behind it.
Formula Composerequations extracted from the text · chained · computed
§1.01 · Definitions
Adjusted EBITDA means EBITDA for such period plus, without duplication, cost savings and synergies (uncapped)…
Eadj = E + ∑i Ai
machine formAdjEBITDA = EBITDA + Σ Addbacks
feeds
§6.04(q) · Investments
…Investments not to exceed the greater of $500,000,000 and 26.0% of Adjusted EBITDA…
B = max( 500, 0.26 Eadj ) $M
machine formBasket = max($500M, 26% × AdjEBITDA)
feeds
§7.11 · Financial covenant
…shall not permit the First Lien Net Leverage Ratio to exceed 3.50 : 1.00…
λ = DnetEadj ≤ 3.50
machine formFLNLR = NetDebt ÷ AdjEBITDA ≤ 3.50×
E · EBITDA$1,840Minput
∑A · Addbacks$260Minput
Dnet · Net debt$6,900Minput
Eadj · Adj EBITDA$2,100M§1.01
B · Basket$546M§6.04(q)
λ · FLNLR3.29ק7.11 · cap 3.50×
⊢ compliant · 3.29× ≤ 3.50× · incremental headroom $450M
Every result remains auditable from input to clause.
Across the portfolio
Find the same risk across the portfolio.
See whether an exposure mechanism is isolated, recurring, or concentrated.
Attention queue
Agreements ranked by material harm and compounding pathways
01
Asteron Consumer HoldingsSenior Secured Credit Agreement · 2024
3 critical findings02
Meridian Energy PartnersFirst-Lien Credit Agreement · 2023
Compounding03
Harborstone MaterialsCredit Facilities Agreement · 2024
2 high findings04
Northline Retail GroupSecured Notes Agreement · 2022
1 critical findingSeverity by agreement
Where harm concentrates across the book
Asteron Consumer2 C · 8 H · 3 M
Meridian Energy1 C · 4 H
Harborstone Materials2 H · 2 M
Northline Retail1 C · 1 H
Hearthland Home1 M
Recurring mechanisms
Exposure patterns repeating across agreements
Collateral release6 / 13
Uptier priming5 / 13
Incremental debt5 / 13
Sacred rights gaps4 / 13
Double-claim structure2 / 13
29 cited findings · 13 agreementsIllustrative data · no live portfolio information
Thirteen agreements arrive as individual documents.
Keep agreed terms and findings together.
Build a searchable record of negotiated terms, protections, connected provisions, and precedent.
See how provisions interact.
Trace LME, covenant, collateral, priming, amendment, voting, and restructuring paths.
Review the source behind each finding.
Inspect the controlling language, connected path, calculations, assumptions, ambiguity, and review points.
Reach a first review sooner.
Bring clause-grounded evidence forward before a financing, amendment deadline, sponsor request, or repricing event.
Keep findings current.
Connect amendments, update selected calculations and protections, and identify changed exposure.
Operational targets
Targets describe intended operating performance and remain subject to workflow, agreement, and evaluation scope.
Internal benchmark · two credit agreements
See the agreement analysis results.
Compare findings shared with DAC's reference set and each system's asserted pathways that failed review.
Benchmark · two credit agreements
See what each system found.
Compare current models across two views: findings shared with DAC's reference set and asserted pathways that failed review.
01 / FindingsPathways and novel mechanics.
See the overlap with DAC's findings across both agreements.
02 / ReviewPathways that failed review.
Inspect blocked, disproved, and defective claims for each system.
Internal benchmark · two credit agreements
Agreement analysis, compared.
Two views of the same benchmark: findings shared with DAC's reference set, and asserted pathways that failed review.
01 / DAC reference overlap
Findings, both agreements.
Bar length shows how many of DAC's 138 findings a system also found: 101 pathways and 37 novel mechanics.
PathwaysNovel
DAC Pipeline138 findings
10137
GPT-6 Astra81 findings
4833
Claude Opus 5.555 findings
2134
Gemini 4 Argon47 findings
1433
Leading LegalTech Platform24 findings
816
GPT-5.6 Sol57 findings
2532
Claude Opus 553 findings
2033
Gemini 3.8 Flash29 findings
1415
Gemini 3.1 Pro Preview15 findings
69
Shared pathwaysNovel mechanics
This view does not count a system's claims outside DAC's set.
02 / Review outcomes
Pathways that fail review.
Bar length shows how many of each system's own asserted pathways were blocked, disproved, or defective.
FailedAsserted
DAC Pipeline0 of 101 asserted
0101
GPT-6 Astra3 of 79 asserted
379
Claude Opus 5.511 of 37 asserted
1137
Gemini 4 Argon18 of 23 asserted
1823
Leading LegalTech Platform14 of 18 asserted
1418
GPT-5.6 Sol8 of 38 asserted
838
Claude Opus 517 of 28 asserted
1728
Gemini 3.8 Flash10 of 17 asserted
1017
Gemini 3.1 Pro Preview4 of 8 asserted
48
DefectiveDisprovedBlocked
DAC's zero refers to its final reviewed output. Other systems were assessed as delivered.
Other scored models currently hidden: GPT-5.6 Sol (57 shared findings; 8 of 38 failed review); Claude Opus 5 (53 shared findings; 17 of 28 failed review); Gemini 3.8 Flash (29 shared findings; 10 of 17 failed review); Gemini 3.1 Pro Preview (15 shared findings; 4 of 8 failed review).
Review the methodology
Agreement set
Coty 2015 first-lien (about $4.5B) and Phreesia 2026 revolver ($275M) credit agreements.
Reference set
DAC's 101 pathways are the reference set by declaration. No human has rated them, so DAC scores 101 / 101 by construction. The set isn't DAC's first pass alone. After the competitors' outputs were frozen, DAC re-checked the pathways they found that it hadn't, and adopted the ones the agreement supports.
Coverage and novelty
The left chart counts how many of DAC's 101 reference pathways each system also found. A separate patterned segment shows how many of DAC's 37 novel mechanics it covered. Novel mechanics are individual mechanics, not full pathways, and Phreesia's 29 have not been validated. A system may assert other pathways that this chart does not credit.
Failure review
The right chart uses a common 0–18 scale. Each failed pathway is counted once, under its worst outcome: blocked by the agreement (per DAC's analysis), disproved by a reviewer, or defective.
Denominators
Failure counts use each system's own asserted pathways: DAC 101, GPT-6 Astra 79, Claude Opus 5.5 37, Gemini 4 Argon 23, and the Leading LegalTech Platform 18. Earlier models, when shown: GPT-5.6 Sol 38, Claude Opus 5 28, Gemini 3.8 Flash 17, Gemini 3.1 Pro Preview 8.
Final-output review
Reviewers flagged 37 DAC statements as defective and none as disproved. DAC corrected or withdrew them, and the corrected text was reviewed again, mostly by one reviewer. Competitor outputs were graded as delivered, with two reviewers per statement. The 0% failure rate describes DAC's final reviewed output, not guaranteed production accuracy.
Discovery context
74 of DAC's 101 pathways rest on a pathway another system in this benchmark surfaced first. DAC re-checked 85 competitor pathways and adopted 81, 69 of them only in corrected form; 2 of those adoptions were later withdrawn. It rejected the other 4. The result reflects the breadth and quality of the combined output, not exclusive first discovery.
Why Fable was not included
Claude Fable 5 was removed from the roster on August 21, 2026, before any model was run, because of an anticipated refusal risk. It was never run on the benchmark, and Claude Opus 5 took its place.
Evidence limitation
Internal DAC benchmark, protocol v1.15, evidence as of October 14, 2026. Review was model-based, with no human adjudication. Some reviewers are also benchmarked systems, and some reviewed their own claims. Four earlier systems are scored in full and hidden by default. Results do not establish production accuracy, legal validity, or quantified cost savings.
ClauseFlow · product preview
View the preview See how ClauseFlow traces a finding to the agreement.
First application · ClauseFlow
See how provisions work together.
Exposure can emerge when provisions interact across an agreement. ClauseFlow traces those interactions, links every finding to the source, and finds recurring exposure across a portfolio.
Explore ClauseFlowFollow connected provisions through the agreement.
Inspect the evidence behind each finding.
Find repeated risk across the portfolio.
Our mission
Empower humanity to command the intelligence it creates.
People set the rules. DAC helps intelligent systems operate within the authority people and institutions define.
Founders
Anish Patel
Founder & CEO
Saw firsthand during the financial crisis how small clauses in agreements could move billions of dollars. Was on the advisory team that supported the issuance of Lehman's final CLO the week before the bank's failure. Leads DAC's product, architecture, and institutional partnerships.
Shalu Maheshwari, Esq.
Co-Founder & General Counsel
Senior IP litigator and strategist turned technology operator. Defines the legal and conceptual architecture that carries agreement logic into engineering and product design while overseeing IP strategy, compliance, and go-to-market frameworks.
Advisors
Dr. Edward Hunter
Director of Engineering, Johns Hopkins DS&AI Institute
AI and computer-vision pioneer whose work in visual tracking and large-scale AI systems helped shape modern motion-capture infrastructure. Advises on technical architecture and long-term AI strategy.
Howard Widra
Former Head of Direct Originations at Apollo; Former CEO and Chairman at MidCap Financial
Private-credit and capital-markets leader who deployed billions in capital and scaled a leading lending platform; brings deep institutional investor relationships.
Wayne C. Plewniak
Managing Director & Head of Fixed Income, GAMCO/Gabelli
Institutional fixed-income leader with deep credit expertise and buy-side relationships.
Jason Turner
Chief Executive Officer, Entanglement, Inc.
Quantum and AI infrastructure entrepreneur; former QED-C (NIST) workforce lead. Advises on scalable AI systems and research partnerships.
Tyler Weiss
VFX Producer, Image Engine
Emmy-nominated producer in game and film visual effects.
VFXGame of Thrones, Avatar: The Last Airbender, Alien: Romulus, 3 Body Problem, and Fantastic Beasts: The Secrets of Dumbledore.
GamingThe Lord of the Rings, FIFA, Madden NFL, Batman: Arkham Asylum, and Deus Ex.
Talk with DAC
Put agreement rules behind AI action.
Tell us where intelligent systems need clear authority, traceable decisions, and human review.
Test ClauseFlow
Test one agreement your team already knows.
Bring one agreement and one known exposure path. ClauseFlow traces the provisions, conditions, and calculations behind it.
Build with DAC
Build the agreement layer.
Whether you are shaping an institution, a consequential AI system, or the next application of DAC's infrastructure, we would like to hear from you.