/hubzoid
HOW IT WORKS · WE BUILD IT, WE RUN IT

We build the AI hub that runs your company.

Our team designs it around how your company actually decides, deploys it inside your perimeter, and runs it for your people. Open-source at the core. Owned by your enterprise.

FOUR LAYERS · COMPOSITION

What the hub is, layer by layer.

01 / INTERFACE
01 / 04

Existing surfaces, no new login.

Telegram, Slack, WhatsApp, web dashboard, in-app buttons, email, Claude Code, IDE. The hub plugs in behind whichever surface a role already uses. No standalone product to roll out.

02 / AGENTS
02 / 04

Role-scoped, three interaction modes.

Tools that run on demand, Q&A that holds the institutional voice, Background workers that ship overnight digests. Each agent carries a memory file, a named owner on your team, and an explicit escalation boundary.

03 / THE HUB
03 / 04

Versioned structured knowledge.

Markdown artifacts that capture decision logic, policy thresholds, ownership maps, escalation routing, vendor terms. Versioned in git, refreshed by utility agents, read by every role-scoped agent at inference time.

04 / CONNECTORS
04 / 04

Read-only by default. MCP-wired.

ERP, POS, drives, chats, accounting. Wired as MCP servers, read-only on deployment. Write access opt-in, audited, reversible. Your stack stays primary; the hub composes on top.

RUNTIME · BUILT FROM PROVEN LAYERS

An open stack. Built on proven layers.

01 / WEB UI
Open WebUI

White-label chat surface that operators see. Customer brand, customer domain, customer SSO. The interactive surface for Q&A and Tool agents.

02 / API BRIDGE
FastAPI · OpenAI-compatible

Translates between the operator surface and the agent runtime. Exposes /v1/chat/completions and /v1/models endpoints so any OpenAI-compatible client can plug in.

03 / AGENT RUNTIME
Agents SDK

Either the OpenAI Agents SDK or the Claude Agent SDK, chosen per the customer's model contract. Same orchestration shape either way. Handles sub-agents, tools, handoffs, and MCP server connections to the customer's existing systems.

04 / MODEL ROUTER
LiteLLM

Routes prompts to whichever model the customer has contracts with: Anthropic, OpenAI, OpenRouter, Bedrock, Vertex. Provider-agnostic by design; no lock-in to a single LLM vendor.

Open-source runtime · MIT licensed · auditable on GitHubView on GitHub

BUILT ON OPEN STANDARDS

PythonFastAPIOpenAI SDKClaude SDKMCPGit
FROM OUR ENGAGEMENTS · WORKING HUBS

What it looks like in production.

Real questions from operators, real answers from agents, pulled from real engagements. Specifics changed for confidentiality; the shape is preserved.

/marketingask · telegram
$ /marketing what should we post for the seasonal promotion this week ask → Theme is set for the week. Last cycle's mid-week posts averaged 2.3x engagement vs launch day. 3 locations are hosting in-store events Friday. Recommend a featured-collection reel shot at the flagship location, pushed to Reels and Stories. $ /marketing slow-mover cadence for next week ask → Featured SKUs rotate down 14 days before each promotion. 5 locations are showing slow-mover signals on the premium range. Recommend a featured-window push next week with the regional brand kit. Want me to brief the studio?
/finance-reportwork with you · slack
$ /finance-report close the books for april report → Closed April. Pulled 2,841 entries from the ledger. Tax workings reconciled. Exceptions: 2 vendor invoices outside policy band, both routed to the regional manager. MIS draft attached. Digest sent to the leadership group. $ /finance-report anomalies in this week's billing report → 3 billing entries flagged. Two duplicate vendor refs. One discount above the auto-approval band. Routed per policy. Awaiting decision before close.
/weekly-reviewwork for you · scheduled
[ scheduled · monday 07:00 · /leadership ] digest → Week 46 leadership digest. Revenue: +8% WoW, pulled from ERP at 06:00. Anomalies: 3 flagged, vendor invoice variance. Q3 close: on track, 2 outstanding approvals. People: 2 attrition risks, ops + marketing. 6 systems read. Audit log: AL-2046. [ scheduled · daily 07:00 · /inventory ] slip → 6 slips processed overnight. 5 cleared. 1 mismatch: ledger 12,400 vs slip 14,200. Posted to the inventory group with slip image and diff. Audit: SLP-2412.
DEPLOY · ONE QUARTER

One quarter to transform. A company that runs on AI agents.

Talk to us