Jaisiu 2.4 · ships with local GGUF support

The AI that runs your fleet.

One CLI. One dashboard. Every model — local GGUF and frontier cloud. Built for sysadmins who ship, not sandbox demos.

curl -fsSL pryzm.at/install.sh | sh
jaisiu — install + first dispatch
$ curl -fsSL pryzm.at/install.sh | sh
✓ Jaisiu 2.4.1 · harness ready
 
$ jaisiu fleet init --model qwen3.5:35b-local
✓ fleet online · 3 workers · model: qwen3.5:35b
 
$ jaisiu ask "why is api latency p99 1.2s?"
● agent: replicator
  investigating 3 traces… found regression
  introduced in commit a3f2e1c
✓ drafted patch · awaiting review
What Jaisiu does for you

A control plane for actual work.

Not a chatbot in a browser tab. An ops-grade fleet you can ship behind your own nginx, on your own boxes, talking to your own models.

Dispatch

One command. Every model.

Route a query to local Qwen, Claude, Haiku, or any GGUF on your bench. Same CLI, same trace format. Cost and latency per model show up in the dashboard before the response lands.

Memory

State that survives restarts.

Sessions, flyweight cache, and quantum-palace recall persist across gateway restarts. Rebuild a node and pick up the same thread — no more "I forgot what we were doing."

Fleet

Workers that fail loud, not silent.

Spawn subagents from the CLI, see them in the topbar pill, steer them mid-run, collapse on completion. Each worker reports its own plan, not a black-box summary.

Pricing

Pay for tokens, not seats.

Pool-based billing means one operator running thirty flows costs the same as one. No per-agent pricing. Capital-positive economics before you hit $110k MRR.

Privacy by design

Your fleet stays yours.

Jaisiu runs on your hardware, behind your nginx. We don't see your queries and we don't train on them.

01

Self-hosted by default

One installer. One process. Listen on a unix socket or localhost; let your reverse proxy handle TLS. No outbound callbacks to pryzm.at unless you opt in.

02

Per-prompt model routing

Classify a query, pick the model on your allowlist, dispatch. Sensitive prompts stay on a local GGUF; only the safe ones ever touch a frontier API.

03

Auditable by default

Every dispatch is a row in quantum palace. Every token is a line in broker-usage. Replay any session, any model, any day. Export to CSV without rate caps.

Simple, transparent pricing

Pick the pool that fits.

All plans include the CLI, the gateway, and every model. You pay for tokens, not seats.

Fleet

For ops running production traffic across hosts.

$149/ mo

Billed monthly · custom pool tuning

  • Harness across 3–5 hosts
  • 4B tokens / mo (pooled)
  • 8 burst slots across hosts
  • Telegram / Slack / GitHub wiring
  • Named support window, 1 business day
Get Fleet

Foundry

On-prem, multi-node, written SLA.

Contact 

From $8,000 / year — quoted per deployment

  • Multi-node Jaisiu on your rack
  • On-prem broker optional
  • White-label option
  • Written SLA, quarterly review
Talk to us

Ship the fleet. Skip the sandbox.

One installer, every model, real ops from day one. Cancel inside the dashboard.

Ready when you are.

Self-host in two minutes. Get Basic, mint a broker key, dispatch your first agent before lunch.

Get Basic

Leave your email.

Get a one-line email when we ship something that matters. No newsletter of the week, no marketing drip.

Talk to us.

Self-host, on-prem, or Foundry with a written SLA — we'll point you at the right tier and help you ship.

Get in touch

We store the minimum data we need. Email signup records your IP, user agent, and the consent text version. Privacy notice · Cookies