“Tell it once. Every AI’s memory.”
A person’s own .said, working inside the AI’s they already use, including Claude, ChatGPT, Gemini and Cursor, over MCP.
| Plan | Price |
|---|---|
| Free | $0 |
| Pro | $10 a month |
| Max | $40 a month |
| Lifetime | $149 once |
| Enterprise | $998 a month |
Every AI is building its own memory, locked inside its own product.
.said is the memory people and companies own: one file, plugged into ChatGPT, Claude, Gemini, Copilot and every AI that speaks MCP, running on a laptop or a company’s own servers, and today on Cloudflare’s edge, built to run on any edge or cloud. Answers cross. Data doesn’t.
¹ said-mcp, November 18, 2026, on a normal computer. Self-reported; reproduction script available to investors.
Every company’s AI has to answer questions that span business units, subsidiaries and partners. Today that means copying everyone’s data into one place first: months of migration, a new database to buy and secure, a new target for attackers, and legal fights over who may see what. .said lets each unit keep its documents where they are and still answer the group’s questions. Only the answer travels.
No migration project. Each unit compiles its own documents on computers it already owns, and the group can ask the same week.
No central copy, no new database, no new target for attackers. Security reviews what already exists.
Every change sits in a hash chain, so the record is tamper-evident. Auditors get proof, not screenshots.
Nothing was copied, so ending a unit's or a partner's access ends it completely.
Why it is urgentData readiness is the top reason enterprise AI projects stall. Through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. Gartner, press release, February 26, 2025
| Workstream | Centralize first | With .said |
|---|---|---|
| Data migration project | Map, move and reconcile every source system | None. Documents stay where they are |
| Vector database | A new platform to procure, run and patch | None. Each unit's signed .said file is its own index |
| GPU embedding | Model compute on every page, again on every change | None. Compiled on ordinary CPUs |
| A second copy of the data | A new store for security to review and defend | None. No copy is made |
| Index freshness | The central index drifts from the source | Each unit recompiles its own file |
Source: SAIDHOME analysis. Qualitative comparison; savings have not yet been measured on a customer project.
| Estimate | Scope | Source |
|---|---|---|
| $52.6B | AI agents market by 2030, up from $7.84B in 2025 (46.3% a year) | MarketsandMarkets, AI Agents Market report, 2025 |
| $48.3B | AI agents market by 2030, up from $8B in 2025 (43.3% a year) | BCC Research, press release, 2025 |
Estimates vary widely between firms. Shown with their source and date.
| Company | Round | Date | Source |
|---|---|---|---|
| Mem0 | $24M seed and Series A | October 2025 | VC Access Online |
| Letta | Seed of $10M | September 2024 | VC Access Online |
| Supermemory | About $3M seed | October 2025 | TechCrunch |
Sources as linked. Shown to indicate investor interest in the category.
Each business unit and partner keeps its documents in one sealed .said file on computers it already owns. An AI anywhere in the group asks; each owner answers at the level it allows. The same file answers from a laptop, a company’s own servers or any edge or cloud network, so one company’s memory serves its whole group, and a person can share their own memory with exactly who they choose: only what is asked crosses, and every share is revocable.
LogEntry 18,442 · underwriting-agent → claims.said · excerpt · signed · previous hash 4be1…9c07
Source: SAIDHOME. Names and places are illustrative.
| Level | What crosses | Example answer | Release |
|---|---|---|---|
| Yes / no | The fact only | Yes. | Owner's policy |
| Pointer | Where to look | Yes: claim CLM-20931, Houston office. | Owner's policy |
| Excerpt | The passage that answers | “Water damage, kitchen, March 2025, paid $14,200.” | Owner's policy |
| Full document | The whole file | Claim file CLM-20931 | A named person approves each release |
Asked: “Has this customer filed a water-damage claim in the last three years?” Illustrative example. Disclosure levels are on the roadmap; today an answer returns the matching passages and their source.
One search across a company’s documents, contracts and source code, in one file. For code, .said understands structure: every function, where it is used and how calls flow. Enterprise AI search either treats code as plain text or hands it to a separate coding tool.
Exact word search in the file, with meaning supplied by the AI you already use. No vector database, no embeddings. When a match is weak, .said asks your own AI to rephrase the question and searches again.
Runs today: the .said engine reads PDF, Word, PowerPoint, Excel, HTML, JSON, Markdown and source code into one file, with code tools for definitions, references, call flow and imports. For comparison: Amazon Q Business GitHub connector (AWS); Microsoft 365 Copilot GitHub connectors, MC1198705; GitHub Copilot vs Amazon Q (Augment Code).
Compliance asks two questions: what could the AI see, and can you prove it? .said answers with a tamper-evident hash chain inside every file, not screenshots. That turns the compliance team from the blocker into the buyer.
Built today inside every .said file: every change sits in a BLAKE3 hash chain, so an edited or deleted line breaks it (said audit --verify). Each AI write is signed with the app's Ed25519 key, and a whole-file seal is optional. Roadmap: signed AI reads (Brazos), and the same signed log for every question and answer at the sharing hub.
Roadmap: each .said file will carry a signed list of exactly what went into it, so a document that is not listed provably could not have been seen. Today an optional seal covers the whole file.
Article 12 asks high-risk AI systems to keep automatic, traceable records. The signed log is designed for that record-keeping. Designed for, not certified.
Nothing was copied, so ending a partner's access ends it completely: no copies to recall, no index to purge.
| Entry | Time (UTC) | Agent | Source | Level | Previous hash |
|---|---|---|---|---|---|
| 18,441 | 14:02:07 | underwriting-agent | claims.said | Yes / no | 7c1e…a410 |
| 18,442 | 14:02:09 | underwriting-agent | claims.said | Excerpt | 4be1…9c07 |
| 18,443 | 14:03:51 | partner-agent | lending.said | Refused | 9f3a…e21d |
Roadmap, illustrative entries. Today the hash chain and signed AI writes live inside each .said file; the hub-level log is next.
| Path | Status |
|---|---|
| claims/2025/ | Compiled · excerpts allowed |
| policies/homeowners/ | Compiled · pointers only |
| medical/ | Not compiled · cannot be disclosed |
| payroll/ | Not compiled · cannot be disclosed |
Roadmap, illustrative. The scope manifest is planned; today the signed seal covers the whole file.
A vector database, GPU embedding and re-indexing every change are the incumbent stack’s biggest recurring bills. .said has none of them, so it costs about 10–40x less to run, and there is nothing to re-index.
Source: SAIDHOME, median per question, no GPU. said-mcp, November 18, 2026, on a normal computer. Self-reported; reproduction script available to investors.
A question over 2,040,352 memories in a 3.46 GB file returns in 21.8 milliseconds (median), no GPU.¹
The laptops and servers already on the books. No new hardware to buy.
Zero embeddings, zero GPUs, $0 vector database: the incumbent stack's biggest recurring costs disappear, so margins stay software-like.
The same file answers from a laptop, a company's own servers, or any edge or cloud network: Cloudflare, AWS, Microsoft Azure, Google Cloud, Alibaba Cloud. It runs today on Cloudflare's edge.
| Bill | Today’s stack | With .said |
|---|---|---|
| Vector database | A new store to run and secure. Pinecone lists $0.33 per GB-month and $16–$27 per million reads, and reads grow with every query | $0. The signed .said file is its own index |
| Embeddings and GPUs | Every page through a model, again on every edit and model change. Embedding APIs list at $0.02–$0.20 per million tokens; an 8x H100 AWS instance is about $55 an hour | $0. Compiled and searched on ordinary CPUs |
| Lock-in and the copy | Vectors are tied to one model and one store, so leaving means re-embedding everything; a second copy of the data to review and defend | None. One file the owner holds, on a laptop, own servers, or any edge or cloud it is deployed to |
List prices on October 17, 2026: Pinecone pricing; OpenAI pricing; AWS p5.48xlarge via Vantage. .said’s own running cost is about $50 a month per business unit, measured on Cloudflare today; its list price is $998.
| Stack | Per month |
|---|---|
| Pinecone Standard + OpenAI text-embedding-3-small | $515 |
| Pinecone Standard + OpenAI text-embedding-3-large | $1,036 |
| Pinecone Enterprise + OpenAI text-embedding-3-large | $1,505 |
| Self-hosted vector database, two nodes, own GPU | $2,085 |
| .said, cost to run, measured on Cloudflare today | About $50 |
Illustrative: one business unit, 1M documents, 1M questions a month, 5% re-embedded monthly, one embedding-model change. Vector stacks: year-1 list prices on October 17, 2026, divided by 12. .said: about 5% of its $998 list price, from its about 95% gross margin measured on Cloudflare today. On Amazon S3 Vectors (December 2025) the vector stack is about $75–$190 a month, so the gap narrows to about 1.5–4x; the embedding pipeline and the build remain.
| Estimate | Scope | Source |
|---|---|---|
| $2.65B | Vector database market in 2025, forecast $8.95B by 2030 | MarketsandMarkets, December 4, 2025 |
| $1.94B | Retrieval-augmented generation market in 2025, forecast $9.86B by 2030 | MarketsandMarkets, 2025 |
| $1.5B | Of $37B enterprise generative AI spend in 2025, spent on storing and retrieving data for LLMs | Menlo Ventures, December 9, 2025 |
Third-party estimates; firms differ widely. Shown with their source and date.
Cost to runAbout 10–40x less to run: about $50 a month to run one business unit on .said against about $500–$2,100 for a Pinecone or self-hosted vector stack. Illustrative, measured on Cloudflare today.
At the $998 list price per business unit we keep most of the running-cost saving as about 95% gross margin at small scale, or we can price below the vector stack to win. Both are open.
Our price is flat; vector reads are not. At 10M questions a month, vector reads alone come to about $110k a year against about $12k for .said.
Vendors quote $30k–$120k+ per production RAG project, 40–50% of it data preparation. That pipeline is what .said removes.
Not yet provenTarget: 4x lower project cost, to be proven on a design-partner project.
Enterprise pricing follows the value that crosses a boundary, not the number of seats.
| Segment | Why they buy | Example |
|---|---|---|
| Groups with acquired or siloed units | Each acquisition adds systems nobody wants to merge | A Midwest insurer that bought three regional carriers asks across all four from day one |
| Banks, insurers and healthcare | Regulated data that cannot legally be pooled | A hospital network answers a payer's question without sending a chart |
| Partner ecosystems | Two companies, two security reviews, one question | A bank and a fintech; a manufacturer in Ohio and its supplier in Michigan |
| M&A due diligence | Diligence without a data room | The buyer's agent asks; the seller's documents stay in the building |
Source: SAIDHOME. Examples are illustrative, not customers.
| Basis | Charged when |
|---|---|
| Platform fee per connected unit | List $998 per connected unit per month, for each business unit or partner connected |
| Introductory pricing | For design partners and first customers, by agreement (for example 50% off the first year); the list price stays |
| Per verified cross-boundary query | Alternative, set with the design partner: charged when a signed answer crosses a boundary |
Source: SAIDHOME. List $998 per connected unit per month; introductory pricing for design partners. Terms are set per contract.
A person’s own .said, working inside the AI’s they already use, including Claude, ChatGPT, Gemini and Cursor, over MCP.
| Plan | Price |
|---|---|
| Free | $0 |
| Pro | $10 a month |
| Max | $40 a month |
| Lifetime | $149 once |
| Enterprise | $998 a month |
Each business unit and partner compiles its own .said. Agents across the group ask, owners decide what leaves, and the log proves it. Priced per verified query or per connected unit.
| Product | Purpose | Includes |
|---|---|---|
| Vault | Legal documents that prove themselves | Capture record, trusted timestamp, signature, retention and legal hold |
| Wallet | Credentials agents can use but never see | Agents receive a handle, never the secret; becoming a passkey manager |
A fixed procedure we run on a call. Each step states what you will see and what it proves.
| Time | Step | What you see | What it proves |
|---|---|---|---|
| 0:00 | Setup | Two laptops, each holding one file: claims.said and partner.said. No shared drive, no cloud copy. | No central copy of the data exists. |
| 0:30 | Ask | An agent asks one question that needs facts from both files. | A cross-boundary question works without pooling data. |
| 1:15 | Answer | Each owner returns only the level its policy allows. The agent combines the two. | Answers cross; documents do not. |
| 1:45 | Audit | Two new signed entries appear in the log. We verify the hash chain in front of you. | The record is tamper-evident. |
| 2:20 | Revoke | The partner's access is revoked in one step. | There is nothing to recall or purge. |
| 2:45 | Ask again | The same question now returns only what the remaining owner allows. | Revocation is immediate and complete. |
Run live by the founder. Timings are approximate.
Pre-launch, working product, self-funded. We have no paying customers or revenue to report yet.
The product is built. The seed proves it inside a paying customer and funds the first enterprise team. Post-money SAFE; expected $14M–$20M pre-money, about 18% dilution; up to $5M only if a lead prices it at $20M or more pre-money.
| Category | Share | Purpose |
|---|---|---|
| Engineering and enterprise team | 41% | Founders, engine and edge engineers, a first enterprise seller: 4 FTE-years in year 1, then 3.75 to month 18 |
| Go-to-market and design partners | 16% | Design-partner pilots, events, content, travel |
| Security, benchmark, counsel | 7% | Independent benchmark reproduction, SOC 2 Type I, pen tests, counsel opinions |
| G&A and infrastructure | 6% | Hosting, tools, accounting, insurance |
| Reserve | 30% | Runway if the Series A slips: with revenue, to about month 26 in the expected case |
Source: staged plan in the SAIDHOME model: year-1 cost lines plus months 13–18 at the expected case’s year-2 rates. Allocations are planned, not committed.
| Scenario | Year 1 | Year 2 | Year 3 |
|---|---|---|---|
| Conservative, seed only | $0.04M | $0.18M | $0.47M |
| Expected, seed only | $0.11M | $0.53M | $1.50M |
| Optimistic, seed only | $0.27M | $1.31M | $3.76M |
| Staged plan, with the $10M Series A | $0.11M | $0.61M | $1.98M |
Bottom-up on today’s list prices: Pro $10, Max $40, Lifetime $149, $1 per agent-day, Enterprise $998 per connected unit per month with 50% introductory pricing in year 1. Expected case: 30k / 90k / 220k sign-ups, 3% paid, 30% churn, 4 / 14 / 36 connected units; the staged plan adds the Series A in year 2 and reaches 55 units. No seed-only scenario breaks even inside three years. Every formula is in the model in the investor pack. Actual results may differ materially.
Next roundNext: a $10M Series A at month 15–18, unlocked by a paying design partner, an independently reproduced benchmark, at least $250k ARR and a named team.
The pack is private and shared one to one, never posted here. Ask for it by email and a person sends it.
Tell us your name, firm, role, whether you are an accredited investor and your typical check size. A person reviews every request and emails you a private link.
Write to investors@saidhome.com with who you are, your fund or angel profile, and what caught your eye. The button below fills it in for you.
Meet the founder, see .said working live, and ask anything: product, market, plan, risks.
Qualified investors get the full model, product and legal documents, and our honest list of risks.
We reply within 2 business days.