TABTAB · LOCAL AI · KNOWLEDGE VAULT · BYO BRAIN
TAB [ˈtæb] — n. acronym, def.
Totally Autonomous Business
The org chart runs itself.

TabTab ships a sealed hardware node running local open-weight models — Qwen, DeepSeek, Llama, Gemma, Mistral, and more — organized into an org chart of AI employees trained on your business knowledge. Prefer a hosted brain? Bring Claude Code or Codex. Same org, your choice of minds.

ORG CHART
HIERARCHY
AGENT ACTIVE

Hire an agent.
Give it a boss.

Every agent gets a title, a reporting line, and a job description. The director delegates work down the org chart. Junior agents escalate up. No flat swarm of bots arguing over a shared prompt — a real hierarchy with real accountability. You approve hires, set budgets, and fire underperformers.

Your agents have structure. Not a chatbot free-for-all.
/DEMO · AGENT VIEW · REV 2026.04LIVE DEMO
console.tabtab.com/agents/social-media-marketing
TABTAB⌘K
Dashboard2 live
Inbox
Tasks
DEPARTMENTS
Sales
Operations
Finance
Marketing
Strategy
AGENTS
Voice Lead Intake
Social Media Marketinglive
Review Harvesterlive
Ad Report
Referral Rebate
Agents Social Media Marketing

Social Media Marketing

AGENT /014 · MARKETING
+ ASSIGN▷ INVOKE⏸ PAUSEACTIVE
Latest RunView details →
succeededb7e2a413Heartbeat
2m ago

5 posts drafted. 2 LinkedIn scheduled. X thread queued for tone review. Brand voice score: 94%. Exiting heartbeat.

Run Activity
Last 14 days
Tasks by Priority
Last 14 days
Critical
High
Medium
Low
Tasks by Status
Last 14 days
Done
Success Rate
Last 14 days
Recent TasksSee All →
TAB-88Draft 5 LinkedIn posts for Q2 product launch
done
TAB-72Schedule Instagram reels for roofing case study
done
TAB-91Tone-match X thread to last 30 brand posts
done
TAB-65Generate ad copy variants for pest control vertical
done
TAB-54Publish weekly engagement report to vault
done
TAB-39A/B test CTA copy on landing page hero
done
TABTAB · DASHBOARD · YOUR COMPANY
ONLINE
5
AGENTS ENABLED
1
TASK IN PROGRESS
$309.14
MONTH SPEND
2
PENDING APPROVALS
RECENT ACTIVITY
2m agoCTOcommittedvault/sops/q2-hiring.md
11m agoCMOscheduledLinkedIn post — Q2 launch announcement
18m agoDispatchcheck-in3 inspections confirmed for tomorrow
34m agoIntakequalifiedlead #847 → estimate scheduled 10:30 AM
1h agoCEOapprovedbudget increase for CMO ($50 → $75/mo)
1h agoCoderdeployedinvoice template v2.1 — 3 files changed
bld 2026.04.22 · sealed · node #0041next heartbeat in 14m
0
Packets left premises
Tailscale · no cloud
71×
Fewer tokens per task
via knowledge subgraphs
24/7
Heartbeat supervision
Agents wake · check · act
§ 00.5 — YOUR COMPANY

Everything to run an autonomous company.

Org charts, budgets, governance, audit trails. TabTab gives your AI workforce the same structure you'd give a human team — then seals it to a machine you own.

Bring Your Own Agent

Any agent, any runtime, one org chart. If it can receive a heartbeat, it's hired.

Goal Alignment

Every task traces back to the company mission. Agents know what to do and why.

Heartbeats

Agents wake on a schedule, check work, and act. Delegation flows up and down the org chart.

Cost Control

Monthly budgets per agent. When they hit the limit, they stop. No runaway costs.

Your Company

One machine, one company. Your org chart, your data, your agents — sealed to a single tenant.

Ticket System

Every conversation traced. Every decision explained. Full tool-call tracing and audit log.

Governance

You're the board. Approve hires, override strategy, pause or terminate any agent at any time.

Org Chart

Hierarchies, roles, reporting lines. Your agents have a boss, a title, and a job description.

Mobile Ready

Monitor and manage your autonomous businesses from anywhere.

§ 00.6 — HEARTBEATS

Agents wake up. Check work. Act.

Every autonomous operation is logged and traceable. No black boxes. You can read the loop.

.tab autonomous operation
ONLINE
[04:15:02] heartbeat.cron → cto.agent: check vault.md
[04:15:04] cto.agent → graphify: subgraph(changes, last 24h)
[04:15:05] graphify → cto.agent: 3 nodes (0.42 KB)
[04:15:07] cto.agent → action: no escalations
[04:15:07] heartbeat.log: 0.003s · $0.00002 · OK
───
[04:30:02] heartbeat.cron → dispatch.agent: check calendar
[04:30:04] dispatch.agent → vault: read schedules/2026-04-24.md
[04:30:05] dispatch.agent → slack: "5 inspections confirmed"
[04:30:06] heartbeat.log: 0.041s · $0.00008 · OK
───
[05:00:00] nightly.index → graphify: reindex vault
[05:00:42] graphify → nightly.index: 14,482 nodes reindexed
[05:00:43] nightly.index → ledger: $0.18 spent this hour
[05:00:43] heartbeat.log: 42.8s · $0.18 · OK
───
[05:15:01] heartbeat.cron → cmo.agent: check content-calendar.md
[05:15:03] cmo.agent → graphify: subgraph(brand-voice → channels)
[05:15:05] cmo.agent → action: draft LinkedIn post (Q2 launch)
[05:15:08] cmo.agent → queue: pending operator approval
[05:15:08] heartbeat.log: 0.072s · $0.0018 · OK · approval_required
4 heartbeats · 0 failures · $0.1820 totalevery action logged and traceable
§ 01.5 — LIVE INTERFACE

Chat, don't script. Talk to your org.

Plain language maps to tickets and tool calls. Your agents respond with cited answers from the knowledge vault. No YAML. No prompts.

TABTAB · YOUR COMPANY
LOCAL
message your org — "CEO, read vault.md and draft Q2 plan"
↳ graph nodes: 1,482 · edges: 6,204bld 2026.04.22 · sealed
§ 01 — THE STACK

Four core systems.
One sealed box.

TabTab is an opinionated assembly of the best open pieces in agentic AI today. Nothing proprietary underneath. You can read every line. You can unplug and take it with you.

ARCHITECTURAL CORE — EVERYTHING ELSE DEPENDS ON THESE
/01
Org Engine
orchestration

Org chart, governance, heartbeats, tickets, budgets. The human control plane for AI labor.

see § 04
/02
Brain Pool
local + hosted

Qwen, DeepSeek, Llama, Gemma, Mistral, Devstral, Phi, Yi on your node. Or bring Claude Code / Codex keys. Swap per-agent.

see § 05
/03
Knowledge Vault
the vault

Your business, in markdown. Dump SOPs, pricing, contacts, playbooks, emails. Your agents read what you write.

see § 02
/04
Graph Index
the wiki / graph

Turns the vault into a queryable knowledge graph. Agents pull compressed subgraphs, not raw files. 71× fewer tokens.

see § 02
§ 05 — THE BRAIN POOL

Your org chart.
Your choice of minds.

TabTab is model-agnostic on purpose. Put a cheap fast model on the intake desk. Put a reasoning heavyweight on the CTO seat. Hire Claude Code when you need a senior engineer. Swap any brain per-agent, per-task, per-budget. The org chart doesn't care.

LOCAL · ON-NODE · ZERO PACKETS OUT
via Ollama · MLX · llama.cpp
Qwen
RECOMMENDED
2.5 / 3 / 3.5 · Coder · Omni
0.5B → 480BApache 2.0
Agentic loops, multilingual, long context
DeepSeek
RECOMMENDED
V3.2 · R1 · Coder V3
16B → 671B (MoE)DeepSeek 2.0
Reasoning, math, algorithmic code
Llama
3.1 · 3.3 · 4 Scout
1B → 405BLlama 3 Community
General-purpose, 10M context on Scout
Gemma
2 · 3 · 4 (MoE)
1B → 27BGemma
Fastest tok/s, safety-tuned, edge-ready
Mistral
RECOMMENDED
Large · Devstral Small 2 · Codestral
7B → 123BApache 2.0
Agentic coding (Devstral), inline completion
Phi
Phi-4 · Phi-3.5
3.8B → 14BMIT
Punches above weight on reasoning
Yi
Yi-Coder · Yi-Lightning
1.5B → 34BApache 2.0
9B matches DeepSeek 33B on HumanEval
Kimi
K2 · K2.5
MoE · ~100B activeModified MIT
Powers Cursor Composer 2
HOSTED · BRING-YOUR-OWN-KEY
keys stay on-node · never relayed
Claude Code
ANTHROPIC
Long refactors, code review, SOP writing
ANTHROPIC_API_KEY
Codex
OPENAI
Rapid generation, broad tool coverage
OPENAI_API_KEY
Cursor / Composer
CURSOR
In-IDE pair programming agent
CURSOR_API_KEY
Aider · OpenCode
OSS
Terminal-first coding agents
any OpenAI-compatible
Runtimes included: Ollama (daily driver), MLX (max tok/s on Apple silicon), llama.cpp (portable). Quantization tuned per model, per node RAM. Minimum node: M4 Pro / 48GB · runs a 14B director + two 7B managers + three 3B operators simultaneously.
§ 02 — TAB · TRAINED ON ALL YOUR BUSINESS

Dump every file.
The agents figure out the rest.

Drop your SOPs, pricing sheets, CRM exports, writing samples, Slack archives, and playbooks into the knowledge vault on your node. The system reads them, extracts concepts, links them into a knowledge graph, and serves the relevant subgraph to whichever agent is on shift. Your agents don't guess — they quote the doc.

[1]Drop files into /vault
[2]System indexes → graph.json
[3]Agent asks → gets subgraph, not soup
[4]Action taken · cited · logged
/VAULT
vault
playbooks/storm-priority.md4.1 KB
writing-style.md2.8 KB
crm/contacts.csv311 KB
pricing-2026.md1.2 KB
crew/rosters.md6.4 KB
legal/terms.pdf88 KB
sops/intake-script.md3.9 KB
insurance/carriers.md5.2 KB
AGENT QUERY · LIVE
? intake.agent → subgraph("storm-priority lead → pricing → dispatch")
↳ 6 nodes, 11 edges returned · 412 tokens (vs 29,440 raw)
SOPSCRMPRICINGVOICECREWINSURANCELEADSDISPATCHESTIMATESCALENDARVENDORSCOMPLIANCEREPORTS
LIVE KNOWLEDGE GRAPH · 1,482 NODES · 6,204 EDGES
§ 04 — HOW THE NODE WORKS

An org chart you can unplug from the wall.

01

Pick your brain

Qwen, DeepSeek, Llama, Gemma, Mistral on-node — or plug in Claude Code / Codex. Different agents can run on different models.

02

Knowledge vault

One folder of markdown is the source of truth. You edit it like Dropbox. Agents read it like Wikipedia.

03

Graph index

Rebuilds the knowledge graph on every commit. Agents query subgraphs — fast, cheap, and cited.

04

Org engine

CEO, CMO, Intake, Dispatch, Coder — hired under one goal, aligned by heartbeat, budgeted to the dollar.

05

Sealed on arrival

Tailscale end-to-end, SSH disabled before shipment, invisible IP. Plug it in and the tunnel comes up alone.

06

Chat, don't script

Talk to your agents like employees. Plain language maps to tickets and tool calls. No YAML.

§ 06 — GET YOUR NODE

Ready to run your own?

Hardware procured new, configured at the bench, sealed, shipped. Three-hour onboarding loads your vault. Weekly office hours. No subscription. No cloud dependency. Ever.

HARDWARE
Apple silicon, brand-new
Tailscale tunnel configured
3-hour onboarding intensive
Lifetime office hours
SOFTWARE
Qwen 2.5 pre-loaded & tuned
Org engine runtime
Knowledge vault scaffolded
Graph indexing daemon