Back to Connectry / The Brain
The company brain for AI agents

Operational substrate
for AI agents.

Most AI deployments bolt onto existing tools and hope for the best. The brain is different. It is the layer beneath: persistent memory, reusable skills, coordinated agents, and live context from every system your company touches. Production-grade. Multi-tenant from day one.

Three Months to six products
Two Coordinating AI servers
36 Domain skill modules
Eight Specialist agents
Layer 1 · Memory

Context that survives every session boundary.

Standard AI has no memory. Every session begins cold. The brain provides a structured, persistent memory layer shared across all agents and sessions. Decisions made yesterday inform the agent you run today.

Memory is not a flat log. It is typed, timestamped, and semantically searchable. Corrections are permanent. Temporary context decays. Architectural decisions live forever with the rationale attached.

Semantic search via LanceDB. Find the relevant context, not just the recent context.
Typed events with TTL: context (7d), task (30d), learning (90d), decision (permanent).
Contradiction detection. New learnings are checked against existing beliefs before writing.
Atomic writes with locking. No race conditions across concurrent sessions.
Memory store · live event log
Decision · permanent
brain.becomes.primary.identity. SF ISV stays as revenue line
2026-05-06
Learning · 90d TTL
Buffer GraphQL has no engagement fields. Engagement requires platform APIs directly.
confidence: high
Correction · permanent
Never use em dashes in external copy. They read as AI slop. Use periods, parens, or restructure.
applied to: all output
Layer 2 · Skills

Reusable knowledge, injected at the right moment.

Skills are structured knowledge modules. Each encodes a domain: how to run a Salesforce security review, how to design an API, how to build a slide deck to brand. Skills do not conflict. They compose.

A skill is loaded at session start when relevant. It is not a system prompt appended blindly. Skills are selected by context, weighted by recency, and promoted automatically when they produce successful outcomes.

36 domain skills in production across Salesforce, development, design, and ops.
Self-promoting: skills that produce clean outcomes get elevated in the rotation.
Skills are versioned markdown. Easy to author, easy to audit, easy to improve.
Cross-agent: a skill written for Archy is immediately available to Clawdy.
Skill registry · 36 modules loaded
  • /sf-security-review Salesforce
  • /connectry-design Design
  • /api-design Development
  • /slide-design Marketing
  • /mcp-development Labs
  • /overnight-build Operations
  • + 30 more skills in rotation
Auto-promoted last session: /sf-security-review
Reason: clean security review output, zero re-tries
Layer 3 · Agents

Specialist agents with distinct identities. One coordinated team.

A single general-purpose model serves no one well. The brain runs eight specialist agents: CEO, Salesforce, Design, Dev, Growth, QA, Research, and Ops. Each has a defined role, a set of skills, an escalation path, and a memory it shares with the rest.

Agents coordinate through a shared event log. When one agent makes a decision, the others see it. When one agent learns something, that learning is available to the whole team. Handoffs are not conversations. They are state transitions.

Eight agents. Each specialised. None general-purpose.
Shared event log for cross-agent coordination. No repeated context.
Model routing by task type. Haiku for lookups. Opus for architecture. Sonnet for execution.
Two server deployment: Archy (primary) and Clawdy (secondary). Cross-machine sync every 5 minutes.
Agent coordination. Archy + Clawdy
Arch
y
Claw
dy
@salesforce-agent, active
Reviewing CH Connector 1.3.1 before submission.
Opus · Security Review skill
@design-agent, active
brain/index.html, scroll storytelling architecture.
Sonnet · connectry-design skill
Shared context synced 3 minutes ago
Layer 4 · Sources

Live context from every tool your company touches.

Context is not copy-pasted. Sources are connected directly via MCP servers: GitHub repositories, Salesforce orgs, calendars, documents, Slack channels, deployment logs. The brain queries sources at session start and when needed. Agents always work with current information.

Sources are normalized into a consistent query interface regardless of origin. GitHub and Salesforce look the same to an agent. That abstraction is the point. Any new source becomes available to every agent immediately.

MCP-native source protocol. New sources plug in without agent changes.
GitHub, Salesforce, Netlify, Buffer, MS365, and custom internal sources.
Knowledge graphs generated per-project. No re-reading files for architecture questions.
Source tokens protected. Immutable flags. Daily guard cron. Backup included.
Connected sources. MCP layer
github. Connectry-io
14 repositories · Knowledge graphs generated
live
salesforce, connectryDevHub
2GP packages · 4 scratch orgs · Security Review status
live
ms365, archy@connectry.io
Calendar · Mail · Teams events
live
netlify · buffer · supabase
Deploy status · Social queue · Intelligence layer
connected
Layer 5 · Cross-machine sync

Two servers. One brain.

The brain runs across two servers: Archy in London, Clawdy in Atlanta. They share one memory, one event log, one knowledge base. Multiple concurrent sessions on separate machines all write to and read from the same substrate.

This is not replication in the database sense. It is a shared understanding maintained through disciplined event emission and a structured handoff protocol. When one agent ends a session, the next agent anywhere in the world picks up with full context.

Shared-context projection cron every 5 minutes. Append-only event log.
SSHFS mounts between servers. Brain files readable from both machines.
Priority event channel for urgent cross-session signals.
Session rotation protocol at 90% context. No work is lost on session boundary.
Cross-machine coordination
ARCHY. London
@salesforce-agent
@design-agent
@ops-agent
5 min sync
CLAWDY. Atlanta
@dev-agent
@growth-agent
@qa-agent
Shared event log, brain/memory/events.jsonl
1,247 events · Append-only · Both servers write
Last sync: 2 minutes ago
What we have proven so far

Most pitches ask you to imagine a product. We are operating ours.

These are not projections or demos. They are what the brain has produced in three months of internal operation.

Six Products built and submitted to Salesforce security review in three months of operation
Three Months from first commit to a full portfolio of AppExchange-grade software
Two Coordinating AI servers in different time zones sharing one operational brain
36 Domain skill modules. Self-promoting on success. No manual curation required.
Eight Specialist agents coordinating through a shared event log without manual intervention
Zero Paying customers yet. Pre-revenue. This is the honest number. The brain is still proving itself.

The Salesforce products are not the point. They are the evidence. The brain produced them. The brain is the company.

The primitives transfer

Salesforce is the proving ground. Not the ceiling.

We chose Salesforce because it is the toughest enterprise stack to ship through. AppExchange security review forced the brain to be rigorous. Named Credentials forced proper auth patterns. Governor limits forced real engineering. The result: memory, skills, agents, and sources that hold under the strictest constraints in enterprise SaaS. Those same primitives run anywhere.

ServiceNow SAP Workday NetSuite Microsoft Dynamics Atlassian

The memory layer does not care what system it reads from. The skills layer does not care what platform the agent acts on. The agent coordination model works the same whether the source is a Salesforce org or a ServiceNow instance. We are not promising integrations. We are describing architecture.

Roadmap

From internal use to multi-tenant SaaS.

The shape of the journey. No specific dates. Just the direction.

Phase 1 · Now
Internal operation

The brain runs in production internally across two servers. Building everything at Connectry. Proving the model works under real enterprise constraints with real deadlines.

Operating
Phase 2 · Next
Architecture documentation and early access

Full architecture documentation published. Early access program for select teams who want to run the brain on their own infrastructure. White-glove setup. Direct founder support.

Coming soon
Phase 3
Multi-tenant SaaS

Hosted brain infrastructure. Teams onboard without managing servers. Skill marketplace. Source connectors as a service. Pricing based on agents, not seats.

Planned
Phase 4
Enterprise deployments

On-premise brain for regulated industries. Dedicated infrastructure. Custom skill development. Integration with enterprise identity providers and existing AI tooling.

Planned
Get notified

Know when access opens.

No pitch deck. No sales call. Just a message when early access is ready.

You are on the list. We will reach out when early access opens.

No credit card. No pressure. We do not share your email.

Privacy Policy

Last updated: February 2026

Terms of Use

Last updated: February 2026