A bit more on Cortex since that's the part most people will be curious about:
The retrieval shape is scoped pgvector cosine over ai_memories and org_brain, joined with the entity graph so a single query like "what's the latest with YC?" pulls every memory, every cortex_event, every CRM record, every email thread, every meeting tied to that entity — across modules, deduplicated, in one shot.
The heartbeat aggregator runs on a schedule and rolls fresh cortex_events into memories: pattern detection, strategic shifts, recurring asks. So the AI doesn't just have raw event logs, it has summarized intent.
Retrieval threshold is per-tool, not global, because confidence requirements for "draft me an email" vs "fire a SAFE" are very different.
All of it is exposed to Claude through MCP. You can plug your own Claude into your StartupStarter workspace and get the same context the in-product S2X assistant has.
I would love to see the real-world hallucination rate of this product. A 2025 report had certain models breaching 80% for the first time ever, and it just keeps getting radically worse.
I'm Jose, founder of StartupStarter. After 5 years building this, I'm putting it in front of you.
Every business runs on a dozen disconnected tools — CRM in one place,cinbox in another, finances in a third, knowledge in a fourth. AI tools have ridden the same trap: a Claude-for-sales, a Claude-for-support, a Claude-for-finance, each one a stateless chatbot you re-onboard every morning. None of them know your business. None of them can do anything beyond chat.
StartupStarter solves both problems with one substrate.
We unified the operational stack — CRM, inbox, finance, agreements, cadences, calendar, knowledge, embeddable site widget — and built Cortex underneath all of it.
Cortex is the company brain. Four core layers:
- org_brain: structured company profile. ICP, product, pricing,
strategic objectives, team, communication preferences. The AI knows
what your company is.
- ai_memories: facts captured across every conversation, categorized
as preference / pattern / strategic / workflow / relationship /
factual. The AI remembers what you told it last week.
- cortex_entities + cortex_edges: canonical people and companies
deduplicated across your CRM, emails, calendar, and threads. The AI
knows who you're talking about, no matter which module surfaces them.
- cortex_events: the activity stream. Every meaningful event in your
business flows through here. A heartbeat job aggregates it into
memories on a schedule.
This is what makes our AI co-pilot (S2X) different from a chatbot.
S2X starts every conversation already knowing who your customers are, what stage every deal is in, what was said in the last email thread, what's on the docket this week, what your strategic priorities are.
And it doesn't just answer. It acts. 163 tools across 16 clusters: it queries deals, drafts and sends emails, builds cap tables, fires cadences, schedules meetings, generates SAFEs, captures leads from your public site widget. 15 inline artifacts render in chat — email drafts, financial dashboards, meeting prep cards, deal triage — that you approve or edit in-place, and the assistant applies.
Cortex is also exposed via MCP. Plug Claude (or anything that speaks MCP) directly into your business. Full read, full write, full context.
This is what most "AI for business" companies haven't built: the substrate.
Frontier models keep getting better. The constant is the value of clean, structured, write-capable data your AI can actually
operate on.
A bit more on Cortex since that's the part most people will be curious about:
The retrieval shape is scoped pgvector cosine over ai_memories and org_brain, joined with the entity graph so a single query like "what's the latest with YC?" pulls every memory, every cortex_event, every CRM record, every email thread, every meeting tied to that entity — across modules, deduplicated, in one shot.
The heartbeat aggregator runs on a schedule and rolls fresh cortex_events into memories: pattern detection, strategic shifts, recurring asks. So the AI doesn't just have raw event logs, it has summarized intent.
Retrieval threshold is per-tool, not global, because confidence requirements for "draft me an email" vs "fire a SAFE" are very different.
All of it is exposed to Claude through MCP. You can plug your own Claude into your StartupStarter workspace and get the same context the in-product S2X assistant has.
I would love to see the real-world hallucination rate of this product. A 2025 report had certain models breaching 80% for the first time ever, and it just keeps getting radically worse.
[flagged]
I'm Jose, founder of StartupStarter. After 5 years building this, I'm putting it in front of you.
Every business runs on a dozen disconnected tools — CRM in one place,cinbox in another, finances in a third, knowledge in a fourth. AI tools have ridden the same trap: a Claude-for-sales, a Claude-for-support, a Claude-for-finance, each one a stateless chatbot you re-onboard every morning. None of them know your business. None of them can do anything beyond chat.
StartupStarter solves both problems with one substrate.
We unified the operational stack — CRM, inbox, finance, agreements, cadences, calendar, knowledge, embeddable site widget — and built Cortex underneath all of it.
Cortex is the company brain. Four core layers:
- org_brain: structured company profile. ICP, product, pricing, strategic objectives, team, communication preferences. The AI knows what your company is.
- ai_memories: facts captured across every conversation, categorized as preference / pattern / strategic / workflow / relationship / factual. The AI remembers what you told it last week.
- cortex_entities + cortex_edges: canonical people and companies deduplicated across your CRM, emails, calendar, and threads. The AI knows who you're talking about, no matter which module surfaces them.
- cortex_events: the activity stream. Every meaningful event in your business flows through here. A heartbeat job aggregates it into memories on a schedule.
This is what makes our AI co-pilot (S2X) different from a chatbot.
S2X starts every conversation already knowing who your customers are, what stage every deal is in, what was said in the last email thread, what's on the docket this week, what your strategic priorities are.
And it doesn't just answer. It acts. 163 tools across 16 clusters: it queries deals, drafts and sends emails, builds cap tables, fires cadences, schedules meetings, generates SAFEs, captures leads from your public site widget. 15 inline artifacts render in chat — email drafts, financial dashboards, meeting prep cards, deal triage — that you approve or edit in-place, and the assistant applies.
Cortex is also exposed via MCP. Plug Claude (or anything that speaks MCP) directly into your business. Full read, full write, full context.
This is what most "AI for business" companies haven't built: the substrate.
Frontier models keep getting better. The constant is the value of clean, structured, write-capable data your AI can actually operate on.
Try it: https://startupstarter.co
Happy to go deep on architecture, Cortex internals, the entity resolution layer, the heartbeat aggregation, or anything else.
— Jose