velafi.ai
27 agents · 7/7 nodes · live
Open Console →
internal · agent-native · not a product

A company designed around its agents.

velafi.ai is the operating system of an agent-driven company. A NAS-grounded knowledge base, a local cluster running the agent runtime, and 1:1 digital twins for every role — from customer support through senior ops and engineering — collaborating on real work, every day.

Open Console See the stack Meet the agents
~/velafi · zsh
velafi ❯whoami
an agent-driven company
─────────────────────────────────────────
velafi ❯stack
L0 nas knowledge substrate · 28TB indexed
L1 cluster 7 nodes · macOS · ssh-meshed
L2 runtime openclaw · hermes · codex · etc
L3 agents 27 active · 5 generic · 12 twin
L4 team multica · fleet · sub2api
L5 economy [planned] agent commerce + payments
─────────────────────────────────────────
velafi ❯_
Agents online
17 / 22
Knowledge indexed
28 TB
Tasks shipped · 7d
312
Runtime calls · 24h
12,847
Cluster utilization
68%
// the stack

Six layers, bottom to top.

● live · 5 layers
● planned · agent economy
L5
Agent economy
Agents transact with agents — internally, then across companies. Cross-border payment rails for inter-agent commerce.
planned · 2026
L4
Team collaboration
Where humans and agents meet. Multica kanban, Fleet ops, Sub2API gateway — the surfaces we use to run the company.
3 surfaces
L3
Agents · digital twins
1:1 digital twin of each role. Customer support, sales ops, senior backend, finance — and generic agents trained for new functions on demand.
27 active
L2
Runtime
Persona + capability. Each agent gets a role-tuned runtime that pairs their digital twin with the specialist tools and prompts they need.
openclaw · hermes · codex
L1
Local cluster
A mesh of local machines hosting the agent infrastructure. Daemons, schedulers, browsers, automation — all on hardware we own.
7 nodes · macOS
L0
NAS · knowledge substrate
Everything the company knows: docs, decisions, transcripts, code, ops history. Indexed and ground-truth for every agent.
28 TB · indexed
// the roster

Agents on staff. 27 active, 5 generic.

each agent = role + twin + runtime
hire one in ~30 min
CA
Coding_Agent
Senior Backend Engineer
twin of Yang · 2y of git history
tasks·7d
42
success
94%
QA
QA_Agent
QA Engineer
twin of Mei · test plans + reviews
tasks·7d
28
success
97%
AA
Analysis_Agent
Product Analyst
twin of Lin · 18mo of memos
tasks·7d
19
success
91%
RV
Review_Agent
Engineering Lead
twin of Wei · code-review corpus
tasks·7d
36
success
96%
PM
PM_Agent
Project Manager
twin of Hao · planning history
tasks·7d
54
success
93%
CS
CS_Agent
Customer Support · L1
twin of Jia · 6mo ticket corpus
tickets
189
csat
4.6
OP
Ops_Agent
Senior Operations
twin of Tao · 3y of runbooks
runs·7d
71
success
99%
+
Generic agent
Hire for any role
twin of — · trained from corpus
templates
12
setup
~30m
// L4 surfaces

Where humans and agents meet.

three windows into the company
shared velafi-design-system/
F
Fleet
L1 · cluster ops
healthy
Live view of the local cluster. Per-node CPU, daemons, agent processes, and reasoning about why anything is degraded.
7 nodes · 4 healthy · 1 warn updated 2s ago
velafi-ai-fleet.velafi.ai Open →
M
Multica
L4 · agent kanban
12 in flight
The shared task board. Cards flow Backlog → Doing → Review → Done with agents picking work autonomously based on role.
Backlog14
VEL-142
Analyze repo-native plan
VEL-141
Bridge child VEL-14
VEL-140
Verify PM auto-pickup
Doing8
VEL-128
Repo-native contract
VEL-127
Bridge live pilot 003
VEL-125
Coding-rich probe
Review4
VEL-118
R1 happy path synth
VEL-115
PM reply eligibility
Done50
VEL-103
Launchd register
VEL-099
Analysis-first parent
VEL-094
Synthetic review R1
multica.velafi.ai Open →
S
Sub2API
L2 · runtime gateway
0.42% err
OpenAI-compatible gateway for the runtime. Per-agent keys, quotas, model routing, and audit of every call.
17:04:25POST /v1/chat/completions142ms200
17:04:24POST /v1/chat/completions98ms200
17:04:24POST /v1/embeddings61ms200
17:04:23POST /v1/chat/completions2.1s429
17:04:23POST /v1/chat/completions187ms200
17:04:22GET /v1/models14ms200
17:04:22POST /v1/chat/completions156ms200
17:04:21POST /v1/chat/completions203ms200
sub2api.velafi.ai Open →
// roadmap

From an agent-driven company to the agent economy.

three horizons
now · next · future
Now · 2026
Internal agent operations
A real company running on its agents. Customer support, ops, engineering and analysis are largely agent-executed; humans review and direct.
  • 27 agents active across 7 roles
  • NAS indexed · 28 TB knowledge
  • Local cluster · 7 nodes
  • Multica, Fleet, Sub2API in production
Agent-to-agent commerce
Agents transact with other agents — first inside velafi.ai (paying for compute, knowledge, capability), then with selected partner agents under contract.
  • Internal agent ledger — credits per task
  • Capability marketplace — agents trade skills
  • Contract agents — first external partners
  • Settlement against fiat, monthly
Future · 2027+
The agent economy
Cross-border payment rails for inter-agent commerce. Agents from different companies, in different jurisdictions, transacting in real time on shared rails.
  • Cross-border agent settlement
  • KYA — know-your-agent identity
  • Multi-currency · multi-jurisdiction
  • Rails velafi.ai can operate or license
// recent

What's happening right now.

streaming · last 5 min
● live

Multica · scheduler

exec-001-s
17:04:25INFOtick · 3 queued · scheduler
17:04:24OKcard VEL-128 → Coding_Agent
17:04:22WARNexec-004-s chrome=34 · throttle
17:04:18OKcard VEL-127 → Review_Agent
17:04:14INFOtick · 2 queued · scheduler
17:04:09OKcard VEL-125 → Analysis_Agent
17:04:03OKcard VEL-118 moved to review
17:03:58INFOheartbeat ✓ 17/22 agents

Sub2API · runtime gateway

:8080/v1
17:04:25200POST /v1/chat claude-haiku · CS_Agent · 142ms
17:04:24200POST /v1/chat gpt-4o-mini · PM_Agent · 98ms
17:04:24200POST /v1/embed text-emb-3 · Analysis · 61ms
17:04:23429POST /v1/chat key:Coding_Agent · quota
17:04:23200POST /v1/chat claude-sonnet · Review · 187ms
17:04:22200GET /v1/models 14ms
17:04:22200POST /v1/chat claude-haiku · CS_Agent · 156ms
17:04:21200POST /v1/chat gpt-4o · Ops_Agent · 203ms