§ 01 ── INTEGRATIONS
One substrate. Above your existing stack.
No re-platforming.
Substrate ingests from your existing tools · not in place of them. Permission-bounded (per F2 Customer Data canonical floor). Real-time on high-signal streams. Schema-normalized to a canonical substrate state via Registry Maintainer (Class B Architectural · T4 Steward). Per Diana Hu's YC framing: provide the models as much context as you would provide an employee. The integration surface is how that happens architecturally.
§ 02 ── PRINCIPLE
We don't replace your stack. We unify it.
Most “consolidation” tools want to replace the tools they consolidate. We don't. Slack stays Slack. Linear stays Linear. GitHub stays GitHub. The substrate sits above all of them · reads from their APIs under permission-bounded access (F2 Customer Data canonical floor) · and reasons across the unified state.
Making the whole company queryable — per Diana Hu's YC framing — requires that every important action produce an artifact the substrate can learn from. The integration surface is what makes that operational.
Disconnection is one click. Aggregates purge within 72 hours (per F2 + F9 cascade discipline). Your team keeps the tools they already love.
SUBSTRATE ABOVE STACK · READ-ONLY · F2 CONSENT-BOUNDED · DISCONNECT 1-CLICK · PURGE 72H
FIG.I1 — Substrate above stack · read-only · F2 consent-bounded permission federation · disconnect 1-click · aggregate purge 72h · per Customization Decision Tree applied to customer source-set selection.
FIG.I1 — Substrate above stack · read-only · F2 consent-bounded permission federation · disconnect 1-click · aggregate purge 72h · per Customization Decision Tree applied to customer source-set selection.
§ 03 ── DEFAULT SOURCES
20 default sources baseline. Subset activated per Customization Decision Tree.
20 default source integrations available baseline for tech-company customers (Series B/C standard pattern). Subset of 10–15 typically active for Seed-Series A (compressed). Long-tail sources added in Phase 1 or Phase 3 per customer pattern · per Customization Decision Tree routing.
20 DEFAULT SOURCES BASELINE · CADENCE ENCODED · L1 TECH-CO PATTERN · CUSTOMIZATION DECISION TREE PER CUSTOMER
FIG.I2 — Twenty default sources · cadence encoded · 10 real-time (webhook) · 7 batch (polled) · 3 on-completion (push) · 20 total baseline. Customer-specific subset per Customization Decision Tree.
FIG.I2 — Twenty default sources · cadence encoded · 10 real-time (webhook) · 7 batch (polled) · 3 on-completion (push) · 20 total baseline. Customer-specific subset per Customization Decision Tree.
This is the concrete tool list Diana Hu names in the YC framing — Slack · Linear · GitHub · Notion · calls · standups — connected end-to-end so the 12-agent canonical fabric (7 Class A operational + 5 Class B architectural) has employee-level context. Long-tail sources (custom DBs · sector-specific tools · proprietary internal platforms) added per customer in Phase 1 or Phase 3. Request an integration in § 08.
§ 04 ── PERMISSION FEDERATION
Three layers. Each one revocable. F2 Customer Data canonical floor enforces.
F2 (Customer Data) is the canonical L8 floor for permission federation: default = nothing · per-category opt-in only · no inherited consent · no default-ON. Cross-category sharing blocked at code-path level. L8-Enforcer (Class B · T4) surveils continuously.
PERMISSION FEDERATION · THREE LAYERS · REVOCABLE · F2 CANONICAL FLOOR · L8-ENFORCER SURVEILLANCE
FIG.I3 — Three layers · source-level (architectural floor · F2 + OAuth scope) · user-level (query-time · Registry Maintainer + Audit Curator) · time-bounded (24h cascade · 72h purge). Cross-page consistency: identical to /security § 04.
FIG.I3 — Three layers · source-level (architectural floor · F2 + OAuth scope) · user-level (query-time · Registry Maintainer + Audit Curator) · time-bounded (24h cascade · 72h purge). Cross-page consistency: identical to /security § 04.
01
SOURCE-LEVEL
Substrate can only access what the connected OAuth scope allows. If you don't grant access to #exec-only · the substrate cannot read it. Source-level scopes are the architectural floor. Configured at Phase 0 per F2 + customer Customization Decision Tree.
02
USER-LEVEL
Queries against the substrate respect the querying user's source-side permissions. You cannot ask the substrate “what's in #exec-only” if you don't have access to it. User-level scopes enforced at query time · permission federation state maintained by Registry Maintainer (Class B · T4 · cross-source health) · enforcement audit-emitted via Audit Curator (F4 · L0 doctrine).
03
TIME-BOUNDED
Revocation cascades within 24 hours. Aggregates derived from revoked data purge within 72 hours. Revocation events audit-emitted at filesystem level (severity:critical NEVER-deletable for F2 floor breach attempts). Per F9 (Employee Privacy) when active for EU / multi-jurisdictional workforce.
§ 05 ── REAL-TIME VS BATCH
High-signal real-time. Low-signal batched.
Sources ingested via webhook (real-time floor) · batch poll (15-min default · configurable) · or on-completion push (Zoom · Meet · Loom transcripts). Index latency floors set so Class A operational agents (T2 drafters) and Class B architectural agents (T4 stewards) operate against fresh substrate state.
INGESTION LATENCY · LOG AXIS · REAL-TIME FLOOR 30S · BATCH FLOOR 5 MIN AFTER 15-MIN POLL
FIG.I4 — Ingestion latency · log axis · real-time floor 30s · batch floor 5 min after 15-min poll · Class A agents operate against fresh state.
FIG.I4 — Ingestion latency · log axis · real-time floor 30s · batch floor 5 min after 15-min poll · Class A agents operate against fresh state.
REAL-TIME · ≤ 30S INDEX FLOOR
High-signal real-time sources — Slack · Linear · GitHub · GitLab · Jira · Notion edits · Datadog · Sentry · Stripe · Vercel · AWS CloudTrail · Zoom / Meet / Loom transcripts on completion — ingested via webhooks or push. Indexed within 30 seconds. Class A operational agents (T2 drafters) operate against this real-time stream: Customer Operations Lead · Decision Brief Drafter · Engineering Steward · Risk & Compliance Lead · Strategy Lead.
BATCH · 15 MIN POLL · ≤ 5 MIN INDEX
Lower-signal batch sources — Google Workspace · Microsoft 365 · Confluence · Figma · Salesforce · HubSpot — polled at configurable intervals (default 15 min · Figma 60 min). Indexed within 5 min of poll. Knowledge Curator (T2 · methodology capture) + Decision Brief Drafter + Strategy Lead operate against the indexed state.
§ 06 ── SCHEMA NORMALIZATION
Native schemas in. Canonical substrate state out.
Each source has its own schema. The Registry Maintainer (Class B Architectural · T4 Steward · canonical role per L1 framework: “agent registry · capability matrix · integration source health · cross-source schema normalization”) maps each native schema to the substrate canonical schema:
Entities (people · teams · projects · decisions · events) · Relationships (cites · blocks · supersedes · routes-to) · Metadata (timestamps · authors · sensitivity tags · freshness · evidence-mode · F2 consent state).
Substrate-level queries operate against the canonical schema. Lineage to the native source is preserved on every entity. Cohen's weighted κ ≥ 0.85 verified by Eval Suite Runner (Class B · T4) on entity extraction across N+ sources. Every integration ingestion event audit-emitted via Audit Curator (F4 · L0 doctrine · NEVER-deletable on critical).
SCHEMA NORMALIZATION · NATIVE → CANONICAL · LINEAGE PRESERVED · F4 AUDIT-EMITTED
FIG.I5 — Native schemas (left) · Registry Maintainer T4 (center · κ ≥ 0.85 verified) · canonical entities (right) · lineage preserved · ingestion events F4 audit-emitted.
FIG.I5 — Native schemas (left) · Registry Maintainer T4 (center · κ ≥ 0.85 verified) · canonical entities (right) · lineage preserved · ingestion events F4 audit-emitted.
§ 07 ── BUILD VS BUY
In-house where it matters. Vendored where it scales.
We build the high-value adapters in-house: Slack · Linear · GitHub · Notion. These are the 80% of value for AI-native companies and they justify the engineering investment. Native event semantics preserved. For the long tail (Salesforce · HubSpot · Stripe · Jira · Confluence · etc.) · we use Merge and Nango under the hood — same canonical API surface to the substrate · faster integration coverage. The adapter swap is invisible to the canonical schema (Registry Maintainer normalizes either way).
IN-HOUSE · DEEP ADAPTERS
- SLACK
- LINEAR
- GITHUB
- NOTION
80% of value for AI-native companies. Justified engineering investment. Native event semantics preserved. Cohen's weighted κ baseline established per adapter at Phase 1 KPI gate.
VENDORED · MERGE / NANGO
- SALESFORCE
- HUBSPOT
- STRIPE
- JIRA
- CONFLUENCE
- & long tail
Same substrate canonical API surface. Faster integration coverage. Adapter swap is invisible to the canonical schema. Per Customization Decision Tree applied to customer pattern.
§ 08 ── REQUEST
Don't see your tool?
Tell us. We prioritize integrations by aggregate customer demand (anonymized · per F-FW1 customer L2 IP isolation). One field in the request: tool name and your email. Custom DB / proprietary API integrations available per Phase 1 or Phase 3 work.
§ ── INTEGRATION STATUS MATRIX
Sources, surfaces, provenance.
The 12 highest-signal commonly-active sources for AI-native tech companies — a subset of the 20-source default baseline (see § 03). All read-by-default · F2 consent-bounded. The other 8 default sources (HubSpot · Salesforce · Stripe · Vercel · CloudTrail · GitLab · Loom · Figma) activate per customer pattern via the Customization Decision Tree. Write surfaces are negotiated per-source at Phase 0 — never on by default.
| SOURCE | SURFACE | LATENCY | PROVENANCE |
|---|---|---|---|
| Slack | ◆ READ | real-time | channel · user · timestamp · thread |
| Linear | ◆ READ | real-time | issue · owner · state · cycle |
| GitHub | ◆ READ | real-time | PR · commit · review · CI status |
| Notion | ◆ READ | near-real-time | page · block · author · version |
| Google Calendar | ◆ READ | real-time | event · attendees · accept-state |
| Gmail / Outlook | ◆ READ | near-real-time | thread · participants · labels |
| Call transcripts | ◆ READ | post-call | meeting · speakers · diarized |
| Figma | ◆ READ | near-real-time | file · frame · editor |
| Salesforce / HubSpot | ◆ READ | real-time | object · field · owner · stage |
| Stripe / Quickbooks | ◆ READ | real-time | txn · entity · ledger account |
| Datadog / Sentry | ◆ READ | real-time | service · alert · trace |
| Custom DBs / APIs | ◆ READ | configurable | schema-normalized to canonical substrate |
→ WRITE SURFACES OFF BY DEFAULT · ENABLED PER SOURCE AT PHASE 0 · D7 §12 STEP 03 VERIFY PRECONDITION · F2 + F-FW3 OPERATOR-TIER REFUSAL-TO-BYPASS
§ 09 ── UNDER THE HOOD
Per-customer source selection + custom adapter requests at Phase 0.
What's public on this integrations page: 20 default sources baseline · cadence + latency floors · 3-layer permission federation (F2 canonical) · schema normalization via Registry Maintainer · build-vs-buy split · write-surfaces-off-by-default. What's available in a 30-min Phase 0 conversation (or under NDA for deeper):
- Customization Decision Tree applied to YOUR source pattern · which subset of 20 default sources activate · which custom long-tail adapters needed for your sector
- Custom adapter pricing + timeline · for proprietary internal platforms or sector-specific tools not in default 20
- Per-source κ baseline · Cohen's weighted entity-extraction accuracy verified by Eval Suite Runner at Phase 1 KPI gate
- Write-surface negotiation · which sources we enable write capability on (auto-send Slack DM · auto-create Linear ticket · auto-comment GitHub PR · etc.) · always per-source human-approver-routing · per F-FW3 architectural floor
- Data residency per source · multi-region customers (EU + US + JP) per MSA § 8.3 + F-FW1 customer L2 IP isolation
- Compliance Mapping Library applied to integrations · GDPR Art. 6 lawful basis per source · HIPAA BAA chain · SOC 2 audit chain integration · sector-specific (PSA · APPI · etc.)
§ END ── CONNECT
If your stack is mostly here — Phase 0 connects it in 1 week (Seed-A) · 2 weeks (Series B/C) · 3 weeks (Series D+). Free. You keep all artifacts.