T▒BL█ 1 ▓░ R█v n▒e b▓ r░g█▒n ███▒ $▓.█M ░▒▓ N▒r█h ▒▓░ E█E▒ ▓░ ███ t▒t█l ▒▓░▒
purpose: financial summarysub-totals + total[financials_q3.pdf, Page 47]
PHOROS AI · ΦΩΣΦΟΡΟΣ
We bring light to the documents AI can't read.
Tables, figures, equations, scanned pages. the most valuable content in your corpus is invisible to ordinary AI search. PHOROS makes all of it legible, reasoned over, and cited with certainty. Powered by COB‑2.
Page-exact citations
Reads tables, charts & scans
Sharp at any corpus size
COB‑2 · flagship engine
See your documents in a new light.
Early access is open. Spots fill in waves.
The blind spot
Most AI search is reading half your documents.
Conventional retrieval treats a document as a wall of text. So the moment real value lives inside a table, a chart, a diagram, or an equation, the system goes blind, then answers you with total confidence anyway. That confident blindness is where hallucinations are born.
Tables become noise
Rows, columns, sub-totals and totals collapse into a soup of numbers with no structure left to query.
Figures & charts vanish
The trend line that decides the answer is, to ordinary AI search, an opaque picture it can neither describe nor find.
Equations turn to gibberish
Mathematical meaning is lost to mangled OCR, so technical corpora answer questions wrong at the worst moments.
No idea where it found it
Without trustworthy, page-exact citations, every answer is a leap of faith you can't audit.
financials_q3.pdf · page 47
scanning
Table
Figure
▓░█ i█▒ge ░▒▓ ███ n▒ r░t█n█i▒n ▒▓░ ███▒ l█n▒ ░▒ q▓▒rt█rs ▓░█▒ ███ ░▒▓
purpose: growth narrative[financials_q3.pdf, Page 47]
Equation
∎ p ~ 1 █ e^▓░(█n▒n░█)▓2^▒t█ ░▒▓ g█rbl█d ▒▓░ O█R ▓░█▒
p ≈ 1 − e−n(n−1)/2t+1
[financials_q3.pdf, Page 47]
replay
The flagship engine
COB‑2
If retrieval-augmented search is a propeller plane, COB‑2 is a fighter jet.
COB‑2 is the engine behind every answer. The moment you upload a document, it makes everything on the page searchable: the tables, the charts, the fine print in the scans. Not just the text.
01
Total ingestion
Every image, table, equation, chart and scan is read and rewritten as faithful, searchable language. If it was on the page, COB can find it and prove it.
02
Context travels with every fact
Every passage remembers where it came from, what it’s for, and how it connects to the rest of the document, so answers land on meaning instead of keyword luck.
03
Deterministic citations
Across a corpus of thousands of pages, every claim points back to the exact document and page it came from. Reliably, every time. The end of "trust me."
Surpasses every document-AI product on the market in accuracy and in the size of library it can master.
COB‑2 is the new standard.
Put COB‑2 on your corpus.
Early access is open. Spots fill in waves.
Inside the engine
How COB‑2 reads everything.
A document goes in opaque and comes out as queryable knowledge. Six stages turn raw pages into answers you can audit.
01
Ingest
You upload. Pages, scans, embedded images, all of it accepted as-is.
02
Translate the visuals
Tables, charts, equations and scans are rewritten as faithful, searchable language.
03
Map the meaning
Every passage is placed on a map of meaning, so ideas match even when the words don’t.
04
Gather
The most promising passages are pulled in, along with their surroundings for context.
05
Weigh
Every candidate is judged for true relevance. Only the best make the cut.
06
Answer & cite
COB commits to one answer, cited to the exact document and page.
Built for scale
Ten documents is a demo. A thousand is Tuesday.
Most document AI demos beautifully on five files and quietly falls apart on five hundred. The bigger the library, the vaguer the answers. PHOROS is engineered the other way around: a multi-tiered routing intelligence works out exactly where in your corpus the answer lives before a single word is written.
Locks on instantly
Name a document, even half-remembered, and the engine scopes straight to it. “The March change order” is enough.
Evidence, not guesswork
Before answering, it weighs real evidence from across your entire library. Actual content, not filenames. Then it commits to the documents that matter.
Sharp at any size
Accuracy that holds from ten documents to ten thousand. Growth in your library never means decay in your answers.
your library · 1,284 documents
sweeping
→ reading structural_specs_vol2.pdf
The way PHOROS answers
Every question runs on Deep Search.
Deep Search Mode is how PHOROS answers. It is the default, and at launch the only mode, because we believe it is the only acceptable one. An intelligent agent plans before it speaks: it sweeps your library for evidence, commits to the documents that matter, interrogates them with focused sub-questions, and refuses to settle for a near-miss. You just ask.
Deep Search · default mode
Deep Search
sweeps the entire library and pinpoints where the answer lives
Deep Search
splits your question into focused sub-queries
COB‑2
answers each one, grounded in the actual page
You
one committed answer, cited to the exact page
Doesn't take no for an answer
If a first pass comes back thin, Deep Search reframes the question and digs again automatically, before you ever see a miss.
Watch it work
Live telemetry streams as it scans, commits and reads. You watch the search happen, not a spinner.
Ask the messy question
No keywords, no prompt tricks. Ask the way you’d ask a colleague. The discipline happens underneath.
Watch Deep Search work on your documents.
Early access is open. Spots fill in waves.
Two ways in
One engine. Two front doors.
COB‑2 powers both of our products. Choose the one built for your world, or talk to us about your own. Building software instead? The same engine is available as an API.
For organizations
ClauseNet
Enterprise document intelligence for legal, finance, construction and manufacturing teams. Upload your library, ask in plain English, and Deep Search returns committed answers with page-exact citations you can defend.
powered by COB‑2
For education
ChatPHD
The same engine, reshaped for the classroom and the LMS. Professors turn course material into a knowledge base their students can actually ask. Cited, accurate, and ready for office hours at 2 a.m.
powered by COB‑2
For developers
The COB‑2 API
One key attaches state-of-the-art document intelligence to your use case. The API is fully managed and built from multiple proprietary PHOROS tools: Chat runs COB‑2 Deep Search, and the Embeddings Database runs on COB‑Index. Three primitives in one, each state of the art, with page-exact citations in every response and low, usage-based rates. No components to tune, nothing to operate. Full documentation lives at cob.dev.
Integration is two calls: one where your customers upload, one where they ask.
POST /v2/chat
Chat · COB‑2 Deep Search
Point customer questions here. At launch, every call runs COB‑2 Deep Search; additional modes become configurable per request later. Answers arrive with citations you can render as links.
POST /v2/documents
Document Storage
Point customer uploads here. Tables, charts, equations and scans become searchable knowledge automatically, with incremental updates and instant deletes. Fully managed by us.
COB‑INDEX · FULLY MANAGED
Embeddings Database
Runs on COB‑Index, our proprietary engine: omniscient over all visible content with full multimodal and OCR support, not just text, and O(log n) query time as your corpus grows. No knobs, no tuning. We run it entirely.
POST https://api.cob.dev/v2/chat { "corpus_id": "kb_7f3a9c", "query": "What retention rate did we agree to in the master subcontract?", "deep_search": true, // launch default: COB-2 Deep Search "citations": "page_exact" }
200 application/json { "answer": "Retention is 10% of each progress payment, reduced to 5% at substantial completion.", "citations": [ { "document": "master_subcontract_2025.pdf", "page": 14 } ], "usage": { "queries": 3, "documents_scanned": 1284 } }
Sample schema for illustration. Get a key at cob.dev, wire up the two endpoints, and your product has a source of truth: document omniscience and chat, attached to your use case. Final spec ships with the docs.
Pick your front door.
ClauseNet and ChatPHD are onboarding now.
Benchmark
The end of the confident guess.
Claims are cheap in AI, so we put COB‑2 up against FinanceBench: an open, public benchmark of real questions over real financial filings. And we grade it the hard way. Every question counts, and a blank answer counts as wrong.
0%
FinanceBench accuracy · full 150-question set
No exclusions. No cherry-picking. Scored on the complete benchmark.
0%
of the page, read
Text, tables, figures, equations and scans. Nothing on the page is left unsearchable.
0
unsourced claims
Deterministic citations tie every statement to an exact document and page.
∞
scale ceiling
From a single folder to a full archive, answers stay just as sharp as the library grows.
Hold us to the number.
See 87.3% accuracy on your own corpus.
About us
Phoros is Greek for the light bearer, the one who carries light into the dark.
ΦΩΣΦΟΡΟΣ · /ˈfo.ros/
That's the whole idea. The knowledge inside your documents has always been there: locked in tables, buried in figures, lost in scans. We built PHOROS AI to carry light to it, so an AI can finally read every page the way the author meant it to be read.
We are a two-person team building something disproportionately ambitious, and we are growing. See what we are hiring for on the Careers page.
Harrison
Co-founder · CTO
Architect of COB. Builds the ingestion engine, the retrieval stack, and everything that makes the page legible.
Zack
Co-founder · CEO
Carries PHOROS to the people who need it. Leads partnerships, growth, and the path to every vertical we serve.
Let there be light on your documents.
Early access is open. Spots fill in waves.
Careers
Come build the light.
PHOROS is a two-person team shipping a disproportionately ambitious product. We move fast, own everything we touch, and measure ourselves against the frontier. If that sounds like home, we should talk.
01
Ship daily
Your work reaches customers the week you write it. No committees, no waiting rooms, no six-month roadmaps between you and impact.
02
Own whole systems
Ingestion to interface, you carry real surface area and the authority that comes with it. Your name is on the architecture.
03
Frontier problems
Retrieval, routing, and reasoning at corpus scales most teams never see. The hard problems are the whole job.
Open role
Founding Engineer
Full-stack, AI-obsessed, allergic to hand-waving. Help us take COB‑2 from breakthrough to industry standard.
Don’t see your role? Convince us we need it: careers@yourcompany.com
Not hiring you yet? Try the product.
Early access is open to everyone.
Legal
Clear terms. No fine-print games.
Plain language wherever the law allows it. Your documents are yours, your data is never for sale, and everything below puts that in writing.
Terms & Conditions · Last updated July 2026
Welcome to ClauseNet (the “Service”). By creating an account or using our AI-powered document intelligence tool, you agree to be bound by these Terms and Conditions. If you do not agree to these terms, do not use the Service.
01Account Creation and Security
Eligibility: You must be at least 13 years of age to create an account or use this Service.
Account Responsibility: You are entirely responsible for maintaining the confidentiality of your login credentials and for all activities that occur under your account. You agree to notify us immediately of any unauthorized use of your account.
02User Content & Warranties
Ownership: You retain full ownership, copyright, and intellectual property rights over any documents, text, or auxiliary files you upload to the platform. We claim no ownership over your data.
Explicit Warranty: By uploading any file, you explicitly warrant and represent that you own the copyright to that material or possess the express legal permission, license, or right from the copyright owner to upload and process it.
Prohibited Content: You may not upload any material that is illegal, defamatory, malicious, violates third-party privacy rights, or breaches our Copyright Policy.
03Acceptable Use & System Integrity
Abuse Restrictions: You agree not to abuse the Service by bypassing rate limits, deploying automated scripts, scrapers, or bots to harvest data, or attempting to disrupt the security, infrastructure, or operational integrity of our networks.
System Overuse: We reserve the right to throttle or temporarily restrict account access if your processing requests disproportionately strain our computational architecture or exceed reasonable usage thresholds.
04AI Processing & Limitations
Nature of AI Outputs: The Service uses artificial intelligence to generate document summaries. Even though the Service is more accurate than its competitors, we do not guarantee the absolute accuracy, completeness, or reliability of any generated summary. Refer to the original materials (documents, files, etc.) before making decisions based on AI outputs.
No Professional Advice: Summaries are generated for informational purposes only. They do not constitute legal, financial, medical, mechanical, or professional advice. You are solely responsible for verifying the critical contents of your original documents before making decisions based on AI outputs.
05Data Storage and Account Deletion
User Control: You maintain absolute control over your files. You may delete your uploaded documents, all auxiliary files, and corresponding chat logs at any time through your account dashboard.
Permanent Removal: Deleting files or explicitly closing your account triggers an automated purge that permanently erases that data from our active databases, subject only to a rolling recovery backup window of up to 30 days (as detailed in our Privacy Policy). We accept no liability for the permanent loss of data that you choose to delete.
06Fees and Billing
Pricing Changes: We reserve the right to introduce fees, modify existing pricing structures, or implement subscription tiers for the Service at any time. Any changes to billing or costs will be communicated to you in advance and will only apply to future billing cycles.
07Termination of Service
Right to Terminate: We reserve the right to suspend or terminate your account at our sole discretion, without prior notice, if you violate these Terms, breach our Copyright Policy (including our Repeat Infringer policy), or engage in activities that threaten the stability, security, or integrity of our systems.
08Limitation of Liability & Disclaimers
“As-Is” Provision: The Service is provided on an “as-is” and “as available” basis without warranties of any kind, either express or implied.
Liability Cap: To the maximum extent permitted by applicable law, our company, its affiliates, and its underlying processors shall not be liable for any indirect, incidental, special, consequential, or punitive damages, including loss of profits, data, or operational capabilities, arising out of your use of or inability to use the Service.
09Dispute Resolution, Governing Law, and Class Action Waiver
Governing Law: These Terms and Conditions shall be governed by and construed in accordance with the laws of the United States, without regard to its conflict of law principles.
Jurisdiction: Any legal action or proceeding arising out of or relating to these Terms must be filed exclusively in the courts located within the United States, and you consent to the personal jurisdiction of such courts.
Class Action Waiver: You agree that any dispute resolution proceedings will be conducted only on an individual basis and not in a class, consolidated, or representative action.
10Changes to the Terms
We may revise and update these Terms and Conditions from time to time. If we implement material updates that impact your operational rights or liabilities, we will notify you via website alert or email before the changes take effect. Your continued use of the platform after the effective date constitutes acceptance of the modified Terms.
11Contact Us
If you have any questions or require legal clarification regarding these Terms, please contact us at support@yourcompany.com.
Privacy Policy · Last updated July 2026
We only collect the absolute minimum required to create your account and process your files. We do not track your activity for marketing or resell your data in any way. Your data and documents belong to you.
01Data We Collect & Why
- Account Data: Email address and login credentials to secure your account.
- User Content: Uploaded source documents, embedding files, and both user queries and COB responses to provide the service.
- Technical Logs: IP address and browser type to prevent server attacks and token abuse.
02Storage & Deletion
- On-Demand Deletion: When you delete a document from your dashboard, it and its corresponding embeddings, queries, and answers are immediately and permanently removed from our active databases.
- Account Closure: If you close your account, all associated metadata, files, embeddings, and related records are permanently deleted.
- Backups: Residual data may persist in encrypted system backups for up to 30 days before being completely overwritten. These backups are strictly used for disaster recovery.
03Data Restrictions & Security
- No AI Training: Your documents and summaries are never used to train, fine-tune, or improve our machine learning models or algorithms.
- No Marketing Trackers: We do not use cross-site tracking cookies or compile behavioral marketing profiles.
- No Third-Party Sales: We never sell, lease, or rent your data. Third-party infrastructure sub-processors handle data under strict confidentiality and zero-retention rules.
- Data Security: We implement industry-standard encryption protocols (SSL/TLS in transit and AES-256 at rest) to protect your files against unauthorized access or disclosure.
04Eligibility & Updates
- Children’s Privacy: Our services are intended solely for users who are 13 years of age or older. We do not knowingly collect personal information from children under 13.
- Policy Updates: We may update this privacy policy from time to time. If we make material changes, we will notify you via a website notification or email before the changes take effect.
05Contact Us
If you have any questions about this policy or wish to request manual data removal, please contact us at support@yourcompany.com.
Copyright & DMCA Policy · Last updated July 2026
We respect the intellectual property rights of others and expect our users to do the same. Because our service relies on user-uploaded content, we maintain a strict policy regarding copyright ownership and compliance with the Digital Millennium Copyright Act (DMCA).
01User Warranties and Account Creation
By creating an account on our platform, you are required to explicitly acknowledge and check a box confirming that:
- You own the copyright to any documents or files you upload to our servers, or;
- You have the express legal right, license, or permission from the copyright owner to upload and process those materials.
We operate under the assumption that you possess these rights for all uploaded content. Uploading material that violates someone else’s copyright is a direct violation of our terms and will result in account termination.
02DMCA Takedown Notices
If you are a copyright owner or an authorized agent thereof and believe that any content hosted on our platform infringes upon your copyright, you may submit a formal DMCA notification. To be effective under the law, your notice must be written and include the following details:
- A physical or electronic signature of a person authorized to act on behalf of the owner of an exclusive right that is allegedly infringed.
- Identification of the copyrighted work claimed to have been infringed.
- Identification of the material that is claimed to be infringing and information reasonably sufficient to permit us to locate the material (e.g., specific user account information, filenames, or unique text content).
- Information reasonably sufficient to permit us to contact you, such as an address, telephone number, and email address.
- A statement that you have a good faith belief that use of the material in the manner complained of is not authorized by the copyright owner, its agent, or the law.
- A statement that the information in the notification is accurate, and under penalty of perjury, that you are authorized to act on behalf of the owner of an exclusive right that is allegedly infringed.
Please send your takedown notice directly to our designated copyright contact:
- Email: copyright@yourcompany.com
- Mailing Address: [Insert Company Physical Address]
We aim to review and process all legally compliant DMCA notices within three (3) to five (5) business days of receipt.
03Content Removal and Counter-Notices
Upon receipt of a valid, compliant DMCA notice, we will expeditiously remove or disable access to the infringing material.
If a user believes their content was removed by mistake or misidentification, they may submit a counter-notification to the email address above containing the elements required by 17 U.S.C. § 512(g)(3). If a valid counter-notice is received, we may restore the content within 10 to 14 business days unless the complaining party files a lawsuit seeking a court order against the user.
04Important Considerations & Enforcement
- Repeat Infringer Policy: We maintain a zero-tolerance policy for intentional copyright infringement. In accordance with the DMCA, we reserve the right to terminate, in appropriate circumstances, the accounts of users who are deemed to be repeat infringers.
- Liability for Misrepresentation: Under Section 512(f) of the DMCA, any person who knowingly materially misrepresents that material or activity is infringing may be subject to liability for damages, including costs and attorneys’ fees incurred by us or our users.
Terms this clear deserve a try.
Early access is open. Spots fill in waves.
Request access
Your documents deserve a better engine.
Tell us a little about your library and we will set you up with early access. Demand is high and we keep capacity matched to the quality we promise, so new customers come aboard in waves. Reserve your place now and we will reach out the moment your spot opens.
Early access open · spots fill in waves
You're on the list.
Thanks! Your spot is reserved. We will be in touch the moment capacity opens, so keep an eye on your inbox.