GuardLabs · demo stand

Multi-Agent AI Coding Server
Claude · GPT · Kimi K3, one gateway

Isolated multi-user sessions for a dev team, each running Claude Code, Codex CLI or Kimi CLI against a single self-hosted gateway — with real-time spend tracking and hard limits per user, per model.

Claude Sonnet 4.5 / Opus 4.5 GPT-5.1 / GPT-5-mini Kimi K3

01 / ARCHITECTURE

One gateway, three model providers

LiteLLM sits between your team and every provider — one API key surface, one place to route, cache, log and rate-limit, regardless of which model each session calls.

Claude GPT Kimi K3 gateway (LiteLLM) → one key, one quota engine, any model per session

02 / ISOLATION

Every developer gets their own sandboxed session

Separate OS users, separate working directories, separate agent process. No shared shell, no cross-reading another dev's files or history — the gateway is the only thing they share.

active

dev1

Backend · Claude Code

Modelsonnet-4.5

Shell useru_dev1

Workdir/home/u_dev1

session bde7d…3a12 · pty isolated

active

dev2

Frontend · Codex CLI

Modelgpt-5.1

Shell useru_dev2

Workdir/home/u_dev2

session 91fa2…7c04 · pty isolated

active

dev3

Data / scripts · Kimi CLI

Modelkimi-k3

Shell useru_dev3

Workdir/home/u_dev3

session 5e12b…f091 · pty isolated

throttled

dev4

Lead / review · Claude Code

Modelopus-4.5

Shell useru_dev4

Workdir/home/u_dev4

session 2c88a…d63e · pty isolated

03 / THE PART THAT MATTERS

Spend & limit monitor — per user, per model, live

Every request is metered at the gateway: tokens, cost and quota usage roll up per session in real time. Cross a limit and that session is throttled or hard-stopped automatically — before it becomes a surprise invoice.

Spend today

$52.62 / $70 cap

↑ 12% vs yesterday

Requests today

1,128

↑ 340 vs 24h avg

Tokens today

7.4M

4.28M in · 3.15M out

Sessions over 90%

2 of 4

⚠ auto-throttle armed

Session	Model	Requests	Tokens in / out	Cost today	Daily limit	Status
D1dev1	claude-sonnet-4.5	342	1.2M / 380K	$14.82	$14.82 / $25.00 · 59%	OK
D2dev2	gpt-5.1-mini	210	640K / 210K	$6.10	$6.10 / $15.00 · 41%	OK
D3dev3	kimi-k3	480	2.1M / 540K	$9.40	$9.40 / $10.00 · 94%	NEAR LIMIT
D4dev4	claude-opus-4.5	96	310K / 95K	$22.30	$22.30 / $20.00 · 111%	STOPPED

gateway rule: hard stop at 100% of daily cap → session receives 429, agent surfaces it in-terminal, no silent overage

updated 4s ago · refresh: 5s

7-day spend, all sessions

daily total, USD

Cost split by session

today, USD