Emerald Coast Systems Group
Our platform

Emerald Coast Systems Group builds open swarm oshal.

oshal is our premier platform: an open-source agent-swarm runtime that accepts work as tickets, routes each phase to a worker bot, and runs those bots against whichever model harness fits the job — on infrastructure the operator owns. Everything below is source you can read or a system you can open in a browser.

The runtime

One platform. Your hardware. Your models.

A swarm controller dispatches phases to worker nodes; each node runs a different agent harness against a different provider, and they coordinate over a message mesh. The controller never calls a model itself — execution, cost capture, and per-user credentials live on the nodes. That separation is what makes the platform portable to an on-prem or air-gapped landscape instead of a vendor's cloud.

Ticket → phase → bot Multi-harness execution Per-user credential isolation Cost captured per call Runs on hardware you own AGPL-3.0 core
Visit oshal.ai → Read the source Browse the app store
Application suites

Five intelligent suites on one runtime.

Applications are packaged, not hardcoded: each is a manifest that contributes its own ticket types, bots, and cockpit surface, and installs from the open store. They are grouped into five suites by who the work serves — the same runtime underneath every one of them.

Intelligent Education

Knowledge work: retrieval over your own corpus, recall across your own history, and interactive formats built on the same engine.

  • World Intelligence
  • Ambient Recall
  • Game Show
  • Kid Lens
  • Intelligent Career

Intelligent Finance

Market research, strategy evaluation, and an autonomous paper-trading desk that publishes its own end-of-session briefing.

  • Intelligent Trades
  • Daily Trade Recap
  • Finance
  • Prediction Markets

Intelligent Office

The daily work surface: mail and messaging triage, document and deck generation, files, identity, and payments.

  • Switchboard
  • AI Office
  • Intelligent Communication
  • Storage
  • Identity Hub
  • Social
  • Payments

Intelligent Home & Life

Everything outside the workday: the connected home, spatial capture, travel and transport, media, and meals.

  • Smart Home
  • Spaces
  • Travel
  • Rides
  • Movies & TV
  • Eats
  • Drone Ops

Intelligent Operations

The engineering suite: incident triage and root-cause analysis, DevOps and secrets, security review, and visual workflow authoring that compiles to a real runtime queue.

  • Intelligent Operations
  • DevOps + Vault
  • Security Center
  • Workflow Studio
  • Bot Forge
  • Cloud
  • Sat Ops

Build your own

The store is open and the packaging format is documented. An operator interview produces a packed single-purpose bot — persona, manifest, and surface — and external agent skills import into the same shape.

  • Open manifest format
  • Packed bots
  • Skill import
Inspect it running

What you can open right now.

Open a real oshal app as a guest — no signup, no card. These run on our own oshal deployment; a guest session is fully interactive for a few hours and then clears itself. Sign in to keep your work and connect your own accounts.

Live · guest

OSHAL Assistant

The natural-language front door to the whole platform. Ask it a question and it classifies the intent, delegates to the right bot, and answers — the same assistant a signed-in operator uses to drive the swarm.

Try it - ask anything · classify, delegate, answer
Open as guest → Sign in
Live data · guest

Intelligent Trades

The Intelligent Finance suite running against a live paper account — the real open book, average cost and unrealized P&L, the strategy sleeves, and the desk's own end-of-session recaps. Paper-traded: no live capital, and nothing here is investment advice.

Posture - paper-traded · real positions · refreshed each minute
Open as guest → See the live book ↓
Live · guest

Intelligent Education

A tutoring suite: a classroom of subject bots that answer questions, ground their explanations in a course corpus, track assignments, and keep a study timeline. Ask the tutor a question and watch a real reasoning turn run.

Try it - ask the tutor · grounded answers · class tools
Open as guest → Sign in
Live · guest

Dungeon Master

An AI game master runs a tabletop adventure — narrating scenes, offering choices, rolling dice, and keeping the campaign timeline. It shows an app owning its own domain: a persona, a store of state, and a cockpit surface over it.

Try it - play a turn · narration · choices · dice
Open as guest → Sign in
Generated artifact

Daily Trade Recap

The desk writes its own end-of-session briefing: results and return since inception, what triggered each position, market sentiment, and the open book — assembled into a deck and narrated, with no human in the loop. This is the platform generating a finished artifact, not a dashboard.

Latest session - July 24, 2026 · narrated video + 9-slide deck
Download July 24 deck 2:17 · archive below
Screenshot of the Cost and Token Race model comparison interface
Live

Cost & Token Race

One prompt, several models, side by side: cost, tokens, latency, and answer quality. This is the platform's provider-neutrality made visible — the same request routed across harnesses so a team can choose a model on evidence instead of habit.

Stack - multi-provider engine · live cost meter
Run the race → Runs on our infrastructure
Diagram of agent workflow controls from intake through routing, review gates, cost logging, and delivery
Source

The platform source & controls

The core runtime is public under AGPL-3.0 and the application store is a separate open repository. The controls deck documents how AI-assisted work is coordinated here — explicit roles, routing, cost visibility, and review gates before anything is called done.

Role-based automation Cost and quality gates Senior-review posture
Evidence - readable source · open manifests · controls deck
Read the source → Controls deck

About these demonstrations. The guest apps open our own oshal deployment; a guest session is interactive for a few hours, then clears. They are prototypes and research systems, not production federal systems. The trading desk is paper-traded — no live capital, and nothing here is investment advice. Our earlier standalone prototypes (voice intake, workforce matching, AI planning) have been retired as their capabilities moved into oshal suites; the browser speech-control agent is currently offline. For a guided technical walkthrough or teaming discussion, contact us →

Live paper book

The autonomous desk, live.

Real open positions from the paper book — average cost and unrealized P&L, refreshed every minute — plus the archive of the desk's generated session recaps. Paper-traded; every fill ties back to the signal that justified it.