Introduction

What Is OmniLink?

OmniLink is a full-stack platform for creating AI-powered agents that can reason, decide, and act. It pairs a cloud-hosted REST API with a lightweight local runtime so that the heavy lifting of large-language-model inference stays in the cloud while the actual tool execution happens on your own hardware — a Raspberry Pi, a laptop, or an industrial gateway.

At its core, OmniLink answers a simple question: How do you let an AI model control physical systems reliably, securely, and affordably?

The answer is a three-part architecture:

1. Cloud AI — Choose from four engines (Gemini, GPT, Grok, Claude) to handle reasoning and decision-making.
2. Local Runtime — The Python library runs on your device. It parses commands, executes tool logic, and reports results back to the cloud.
3. Transport Layer — REST and HTTP bridge connect the two halves with minimal latency.

Core Value Proposition

Cloud-Orchestrated AI + Local Tool Execution = Minimal API Costs

Traditional approaches to agentic AI send every intermediate step back to the model. Each round trip costs tokens. OmniLink inverts the pattern: the cloud AI makes a single decision, the local runtime executes it, and only the final result is sent back for analysis.

This 1-credit architecture means that most interactions cost exactly one credit to start a tool run and one credit for the final analysis. Everything in between — state polling, action execution, memory persistence — runs locally at zero marginal cost.

Why This Matters

Key Differentiators

Multi-Model Support

OmniLink is not locked to a single model provider. Four AI engines are available out of the box, each routed through a unified API:

• g1-engine — Gemini (fast, cost-efficient)
• g2-engine — GPT (high accuracy, complex reasoning)
• g3-engine — Grok (real-time knowledge)
• g4-engine — Claude (nuanced instruction following)

Switch engines per request or per agent — no code changes required.

Command Parsing Engine

The OmniLinkEngine provides deterministic command matching using a template syntax with typed variables. This means your agent can handle both natural-language input (via the cloud AI) and structured commands (via local parsing) in the same pipeline.

from omnilink import OmniLinkEngine

engine = OmniLinkEngine()

@engine.command("move [direction:str] [distance:int] meters")
def move(direction: str, distance: int):
    """Move the robot in the given direction."""
    return f"Moving {direction} by {distance}m"

Transport Bridges

Two built-in bridges let external systems interact with your agents without writing custom networking code:

• HTTP Bridge — Exposes a REST API on port 8080. Send commands via POST /command.
• HTTP Bridge Endpoints — Exposes /command, /feedback, and /context endpoints at localhost:8080.

Knowledge Integration

The cloud knowledge upload feature was removed. Agents now read knowledge from per-agent local folders (agents/<name>/knowledge/) checked into the agent repo and searched from Python via the search_knowledge tool.

Who Is OmniLink For?

OmniLink is built for developers and teams who need AI agents that do more than chat. If your project involves any of the following, OmniLink is a strong fit:

• Robotics — Autonomous navigation, mission planning, sensor fusion, and command dispatch for mobile robots and manipulators.
• IoT & Smart Home — Control lights, locks, thermostats, and sensors through natural language or scheduled routines.
• Industrial Automation — Gateway-level command orchestration, sensor monitoring, and alerting on factory floors.
• Customer Support — Knowledge-grounded agents that handle multi-turn conversations with memory and context.
• Education & Research — Rapid prototyping of agent architectures without managing model infrastructure.

Platform Components

Python Library

Install with pip install omnilink (or pip install -e omnilink-lib/ from the repo). The library contains:

• OmniLinkEngine — Command parser with template matching and variable extraction
• ToolRunner — Cloud-orchestrated local tool control
• OmniLinkClient — REST API client for all platform endpoints
• OmniLinkHTTPBridge — REST bridge on port 8080

REST API

The API lives at https://www.omnilink-agents.com. Key endpoints include:

Dashboard

The web-based Dashboard is your control center. From here you can:

• Configure agents (system instructions, persona, available commands)
• Create and switch between agent profiles
• Adjust voice settings (STT/TTS providers and voices)
• Select AI engines and operational modes
• Monitor usage, manage plans, and view billing
• Configure agent connections and view the Heart visualization
• Browse logs and manage active ToolRunner runs

The 1-Credit Architecture

OmniLink’s cost model is designed around a simple idea: pay for decisions, not for execution.

When a ToolRunner session begins, the cloud AI analyzes the user’s request and issues a kickoff command — that costs 1 credit. The local runtime then enters a main loop: it polls state, executes actions, updates memory, and checks for new UI commands. None of these local steps cost anything.

When the task completes (or a periodic review is triggered), the results are sent back to the cloud for a final analysis — another 1 credit. The total for even a complex, multi-step automation is typically 2 credits.

# Typical ToolRunner credit flow
#
# 1 credit  ──▶  Kickoff (cloud AI decides what to do)
# 0 credits ──▶  Local loop: poll, act, persist, repeat
# 1 credit  ──▶  Final analysis (cloud AI summarizes results)
# ─────────
# 2 credits total

Compare this with a conventional agent loop that calls the model at every step. A 10-step task costs 10+ API calls in a traditional setup; in OmniLink it still costs 2.

Next Steps

Ready to get started? Here is the recommended reading order:

1. Quick Overview — A guided tour of the Dashboard and platform features.
2. Use Cases — See how OmniLink is used in robotics, IoT, support, and more.
3. Quickstart — Install the library and send your first command in under 5 minutes.
4. Architecture — Understand the full system design, data flow, and security model.