context layer
primary · os team · august 2026
primary · os team · august 2026

The Context
Layer

One shared map of the firm — people, companies, investments, funds, impact — that our team and our agents both build on.

person company investment fund impact
outline
01 · ontology
Map what exists
The firm's objects and how they relate.
02 · schema
Set the rules
Deterministic guardrails for agents to work within.
03 · ingest
Build the network
Email & calendar participants and LinkedIn connections are automatically ingested to build and maintain the network graph.
04 · sensitive data
Handled by agents
Agents screen your more sensitive data, so nothing personal or unnecessary lands in the layer.
05 · maintenance
Keep it trustworthy
Owned, reviewed, versioned — like code.
the stack for reference
Lovelace
Central
unified UI
your agents
built on top · via API + MCP
The Context LayerNetwork · people · companies
they build & maintain it · ✓ your approval for anything private
Ingestautomated
participants & metadata, deterministically
Cloud agent platformFleet
read, enrich & decide what's worth sharing
Both build & maintain the Context Layer — and both can use your connected third-party tools.
EmailCalendarSlackGranolaLinkedInAffinityNotion
your tools & third-party sources
01ontology 02schema 03ingest 04sensitive data 05maintenance

The lifecycle of an investment

And everything related to that lifecycle, we aim to represent in our ontology.

INVESTMENT LEAD * affinity · lovelace OFFER TERM SHEET meeting * calendar · email forward-deployed impact INCUBATION PARTNER INVESTS TERM SHEET SIGNED Carta cap table → portco INVESTMENT ∈ fund PORTFOLIO CO IMPACT impact teams GTM talent design strategic finance impact / embedded work * = where the data comes from
01ontology 02schema 03ingest 04sensitive data 05maintenance

The object map.

We are mapping the timeline from the last slide to an ontology — a sample of some familiar objects and their links.

person

investment_opportunityfuture_founderportco_employeelp
src · gmail · calendar · pdl · linkedin

company

portfolio_companypotential_investmentseen_deal
src · affinity · pdl · carta

investment

The record that a deal happened. An LP's commitment is an investment of type fund_equity.

src · carta · affinity

fund

A vehicle the firm raises and deploys — e.g. Fund 6. SPVs and select funds are funds too.

fund ⟶ [ investment ]

impact

engagementembedded_project

An engagement is the agreement between a portfolio company and an impact team.

the ontology · requirements

What an agent-ready ontology needs

Eight capabilities, abstracted from Palantir's AIP — and where Central, our hand-built version, already stands.

connect

Pull from systems of record on a schedule, with provenance. Partial

clean

Dedupe, resolve entities, enrich at ingest. Strong

define

Objects, properties, typed links — in plain language. Yes (in code)

hydrate

Keep objects live, not a query someone re-runs. Yes

act

Every write a typed, audited action. Half

govern

Permissions enforced in the data layer. Weak

agents

Schema as the map; search, read, act as tools. Reader, not writer

evaluate

Golden evals gate changes; audit every call. Yes

Central already has the hard parts: resolved entities, fact-level provenance, reviewed writes, evals. The gaps are increments — let the agent propose changes, add write-back & notifications, add semantic search, expose an MCP surface, tighten permissions. Not a rebuild.

01ontology 02schema 03ingest 04sensitive data 05maintenance

Schema constraints or rules

Deterministic guardrails for agents — lines they can't cross.

example

A company becomes a portfolio company only once its Carta cap table exists.

company.type = portfolio_company  ⇒  carta_cap_table.exists
object a thing we keep records on link how two objects connect action a change of state, sometimes gated by approval function logic that computes, never hand-edits data source the real system it comes from
01ontology 02schema 03ingest 04sensitive data 05maintenance

The schema is an input.

Palantir likes to think of the schema as data, not code — so the platform around it doesn't change when the schema does. It's built to evolve, and agents are how it evolves.

schema agents watch identifies patches or improvements requests you approve schema evolves
  • I expect some of the first agents we deploy as a team to be tasked with exactly this — maintaining the schema and proactively proposing improvements and adjustments from what they see in the data.
theo

Incubations, live.

01ontology 02schema 03ingest 04sensitive data 05maintenance

Ingest builds and maintains the network graph.

Software, not judgment. Every email, calendar event, and LinkedIn connection adds people, companies, and the links between them — from the participants and metadata only. Subjects and bodies are never read or stored. It's also what finally lets us count the meetings we source.

EMAIL participants CALENDAR EVENT participants participants & metadata only subjects & bodies are never ingested person company from ingested participants
01ontology 02schema 03ingest 04sensitive data 05maintenance

Agents run on your behalf.

With a subset of your permissions — which MCPs, which tools within them, and to what degree.

FFleetworkspace / connectors
your connectors — the ceiling
Af
Affinity
deals, people, lists
12 tools
G
Gmail & Calendar
read, draft, schedule
connected
Cl
Clay
enrichment
3 tools
+ add connector · any remote MCP server
FSourcing agentrequest · just now
JJordan Foxcreator · OS team9:14 AM
I built this agent and think it would help you. Want it to run on your behalf? It only uses what you allow below.
Agents run as you, so please review the permissions it has.
Af
Affinity
read deals & people
always
Cl
Clay
enrich a new contact
ask first
G
Gmail
not granted
off
Run on my behalf Not now
Built by the OS team · runs with your access, never more.

Unlocking multiplayer.

You don't have to build your own agents, and your data doesn't have to leave your hands. The OS team builds the Fleet; you accept an agent; it runs on your data, for you.

os team builds agent you accept your data
permissions

Everyone's access, no friction

Multiplayer navigates everyone's permissions for us — removing the friction those permissions usually introduce when the team is solving problems for the firm.

leverage

The whole firm behind each agent

An agent the OS team builds once runs for everyone, and each one works from the firm's shared context — not just what one person happens to know.

Context outside of our infra

The data can stay where it is

Every tool was built to hold one kind of work well — comments in Docs, threads in Slack, pages in Notion. We don't fight it. The layer keeps a reference to where context lives, never a copy.

Docs · Slack · Notion · Affinity stay put the Context Layer keeps a reference
reference · our route

No copy to sync. The source stays the source of truth — permissions included.

bring it all in

You own freshness forever, and two systems to keep healthy instead of one.

TBDhow we search into it — buy (Glean) · hybrid (vector DB) · build (Elastic + embeddings)

01ontology 02schema 03ingest 04sensitive data 05maintenance

Context is the new code.

Every change is owned, reviewed, and versioned — so an update never quietly breaks someone's workflow.

Add 3 people, 1 company to the networkaccepted
Proposed by Sourcing agent · reviewed by owner
+ person · Maya Chen → Rune+ signal · Halyard left stealth
owner · Jordan Foxmerged
Reclassify Sferic as portfolio_companydeclined
Blocked by the schema — no cap table on record yet.
× company.type = portfolio_company requires carta_cap_table.exists
owner · Theo Sunnot merged
public

Anyone can use it, and anyone can propose an improvement.

owned

Read or run it freely; changing it goes through the owner.

private

Agents on your own data are yours alone. Sharing is your call.

for discussion

The calls we need to make.

Search into the layer — buy or build?
Embeddings + retrieval, so agents can find the right context.
Glean · meeting their reps open-source alt · exploring home-built · Elastic + a cloud embeddings API
trade-off: buy = fast, less to run · build = control, spend on tokens not seats
The CRM & the network — buy or build?
Where people, companies, and deals live.
off-the-shelf CRM Twenty · trialing homemade · Supabase
trade-off: buy = proven & supported · build = first-party IP, we own upkeep — we lean build (Lovelace/Central pattern)
Token budget — the trade behind every call.
Running agents seriously is expensive — that's not a bug, it's the cost of doing business in this era.
default to open source buy only where it clearly pays put the savings into tokens
We're a small team, so normally I'd always recommend we buy everything we can — but I want us to keep as much money as possible for tokens. So instead I'm recommending we neither build nor buy: we leverage open-source software as much as we can.

Exploring now · graph databases for the network, and self-hosting our own fork of Twenty (open-source CRM) — concrete directions on the first two calls.

1 / 22