DUBHE Co. Ltd.

Alioth AI Platform™

Overview

An air-gapped AI development platform seamlessly connecting task analysis, context generation, and AI agent coding—Planner generates verified Context, and Coder generates code strictly based on that verified Context.

Automated Prompting AI Dev Tool

Planner : Document analysis, requirement breakdown, part-by-part Context finalization
Coder : Confirmed Context-based Agentic Coding
Context Protocol : Planner ↔ Coder link, maintaining Context consistency

Atomic Task Delegation & Hallucination-Minimizing Architecture

Session Key Issuance : 1 Screen = 1 Package
Information Delivery : Essential information delivered, pointer guides for references
Feedback Reports : 4 report types, automated feedback into Planner

Air-Gapped & Open LLM Operations for Absolute Security

LLM Operations : On-Premise environment, selection and operation of OpenSource LLMs
Security : Fundamentally blocks external data leaks and security incidents

sLLM Optimization & Harness Engineering

Performance Optimization : sLLM-based, Harness Engineering applied
LLM Operations : Open LLMs tailored to client enterprise environments
Model Selection : Gemma, Chat OSS, Kimi, etc.

Zero Worry of Data Leaks

Alioth AI Platform

Direct installation on internal corporate portals fundamentally blocks the risk of data and source code leaks.

Alioth AI

Direct Internal Portal Installation

None(Fully Air-Gapped)

None

Multi-LLM Routing Support (Select & switch models as needed)

No Additional Fees Post-Install (Initial setup-focused)

Optimized

Not Required

Deployment Method

External Data Transmission

Source Code Leakage Risk

Model Architecture

Cost Structure

Financial Security Standards

Internet Connection Required

Public Cloud AI

Cloud-Based

Required(Sent to external servers)

Possible Risk

Fixed Single Model (Locked to proprietary model)

User/Token-Based Billing (Costs rise with users)

Separate Review Required

Required

Deployment Method

Alioth AI

Direct Internal Portal Installation

Public Cloud AI

Cloud-Based

External Data Transmission

Alioth AI

Zero external data transmission (Internal air-gapped network)

Public Cloud AI

Inputs and outputs transmitted externally

Source Code Leakage Risk

Alioth AI

Zero security risk (Fundamentally blocked)

Public Cloud AI

Risk of source code leaks remains

Model Architecture

Alioth AI

Multi-LLM switching support (Select & switch models as needed)

Public Cloud AI

Locked to single proprietary model

Cost Structure

Alioth AI

No additional usage fees post-installation (Initial setup-focused)

Public Cloud AI

Continuous pay-as-you-go usage billing

Financial Security Standards

Alioth AI

Fully optimized for financial compliance & security standards

Public Cloud AI

Risk of falling short of compliance standards

Internet Connection Required

Alioth AI

Not required (Fully functional in 100% offline air-gapped networks)

Public Cloud AI

Continuous high-speed Internet connection required

Planner ↔ Coder Dual-Engine Integration

Current Primary Verified & Applied

On-Premise OpenSource LLM Operations

Freely select and operate OpenSource LLMs within the client's internal network, fundamentally blocking external information leaks and security risks.

Optimal Model Selection & Failover per Task

Designate optimal models tailored for specific analytical tasks, with automated failover to standby models ensuring continuous AI reliability.

Verified-Only Models in Production

New models are vetted through rigorous quality verification before production deployment, safely replacing legacy models as needed.

Multi-LLM Routing

Confirmed Gate

Common Assets & Rules

Feedback Loop

Audit Trail & History

Alioth Planner

1 Screen = 1 Package = 1 Session Key

・Common Context :Project-wide global rules
・Screen Context :Detailed screen-level specifications
・Reference Info :On-demand reference materials

Alioth Coder

Dev Directives

Coding Plan

Code Generation

Verify & Review

Reporting

Alioth Planner

Select and operate optimal LLMs based on verified Context and governance frameworks, managing security, quality verification, audits, and model transitions in a single unified web workspace.

Alioth Planner 9 KEY FEATURES

Key Features

CONFIRMED GATE

Confirmed-Only Delivery to AI

Only verified requirements passing the gate are authorized into AI tasks

Unconfirmed draft specs are structurally blocked from AI execution

JOB SPLITTER

Unified Single Source of Truth

Centrally manage source specifications of requirements, screens, and APIs in Planner

Code changes are reflected strictly through feedback reports

3-LAYER CONTEXT

Essential-Only Context Delivery

Deliver only confirmed rules and specs, guiding other references via file paths

Eliminate context noise to drastically minimize AI hallucinations and costs

SESSION CONTROL

AI Work Scope Isolation

Component-specific AI touches only its designated files

Structurally prevents unintentional modifications in other areas

CONSTITUTION

Constant Project Rule Enforcement

Apply identical architectural conventions and versions across all AI tasks

Omitted rules are updated into subsequent versions via feedback reports

FEEDBACK LOOP

Automated Work Result Feedback

Automatically retrieve and integrate common assets, actual code, and improvements

A self-refining iterative loop that matures with every iteration

AUDIT TRAIL

Automated Audit Trail Preservation

Automatically record requirement edits, confirmation timestamps, and AI prompt logs

Automated generation of audit trails for enterprise compliance

PROJECTION

Automated Project Documentation

Automatically generate deliverables, progress rates, and WBS from raw work logs

Minimizes overhead for manual reporting and documentation

WORKSPACE

Unified Collaborative Web Workspace

Seamlessly share analysis, confirmations, and progress among PM, design, and dev teams

Dedicated administrative back-office included

Alioth Coder

Provides an end-to-end air-gapped Agentic Coding environment utilizing client LLMs, ensuring enterprise code quality via harness-based verification and governance.

Alioth Coder 9 KEY FEATURES

Key Features

JOB SPLITTER

Intelligent Task Subdivision

Subdivides development prompts into manageable units for stable LLM processing

Uninterrupted continuous execution of heavy tasks

KNOWLEDGE DATA

Selective Knowledge Injection

Manages dev & secure coding guides as .KD files, selectively injecting per stage

Strict enforcement of corporate standards during code generation

PROMPT COMPACTOR

Full-Pipeline Context Compaction

Overcomes small-model limitations via end-to-end prompt and context compaction

Sustains long-running workloads on resource-constrained GPUs

QUALITY GUARD

Automated Quality & Security Audits

Automated inspection of code verification, peer review, and build outputs

Vulnerability audits compliant with strict financial & security standards

HISTORY MANAGER

Granular History Management

Dual local-server recording of work trajectories across the entire pipeline

Guarantees reproducibility and provides authentic audit trails

AIR-GAPPED

Air-Gapped Offline Completeness

Fully operational in network-isolated setups via self-contained offline VSIX installer

Distributes VSIX & CLI packages via corporate internal portals

EXTENSION & CLI

Identical Quality Across 2 Channels

Shares the same execution Harness across interactive Extension and automated CLI

Unattended CLI-based overnight batch execution for massive tasks

UNDO & CHECKPOINT

Robust Change Control & Rollbacks

Accept/Undo changes per modified file and execute batch rollbacks via checkpoints

3-tier permission governance based on Allow · Deny · Ask

BATCH SCHEDULER

Heavy Batch Task Scheduling

Schedule heavy batch jobs for off-hours and track real-time execution status

Comprehensive retry history logs and reporting for failed tasks