AlphaOne AI
AI CODING WORKBENCH • RELEASE CANDIDATE

AlphaOne Harness

powered by AlphaOne HyperRouter

A workspace-aware AI coding workbench designed to understand projects, prepare implementation plans, edit source code, build and test, diagnose failures, repair and re-validate changes, review AI edits, and finish with an evidence-grounded work result.

AI model connectivity is fixed to AlphaOne HyperRouter; provider credentials and physical model routing remain managed by HyperRouter.
AlphaOne Harness Desktop Overview

From Workspace Understanding to Reviewed Results

AlphaOne Harness combines a familiar development workbench with a governed AI Agent workflow. The application owns the coding experience; AlphaOne HyperRouter owns model connectivity, provider routing, Local AI, quota handling, and Auto RollOver.

AlphaOne Harness Agent workflow, workspace intelligence, editing, terminal, build/test, review and developer control
AlphaOne HyperRouter Provider/model routing, Local AI, BYOK cloud access, Auto RollOver and compatibility
AI Models Local Device GGUF, Local Cloud, Direct BYOK providers and HyperRouter fallback routing

Core AlphaOne Harness Features

These capabilities are part of the completed AOH Release Candidate path. Future roadmap items are shown separately below.

Workspace & Project Intelligence

Discover project structure, resolve the relevant project or module, locate likely source files, and choose appropriate build/test targets without requiring the user to provide every exact path.

Implementation Plan & Progress

Present implementation plans for review, accept multiple feedback items, request revisions, authorize implementation, and surface structured progress during the Agent lifecycle.

Integrated Code Editing

Multi-document editor, workspace Explorer, file lifecycle, Find/Replace, save protection, source selections, and AI actions including Explain, Fix, Improve, and Ask.

Build / Test / Diagnose / Fix

Run project-aware build and test operations, inspect structured diagnostics and raw output, repair failures, rebuild, and re-test with bounded validation behavior.

AI Diff Review

Review tracked Agent changes in the workbench with file and hunk-level controls, including explicit Accept and safe Reject workflows before changes are considered approved.

Checkpoints, Undo & Redo

Agent checkpoints and restoration workflows let users undo an Agent run, restore earlier checkpoint state, and redo supported restored work while preserving review history.

Terminal, Output & Problems

Integrated interactive terminal plus Output channels, Problems aggregation, build/test diagnostics, logs, and developer-facing execution evidence.

Approval & Sandbox Controls

Govern Agent operations through centralized approval and permission policy while retaining sandbox and tool-execution boundaries as authoritative controls.

Dynamic Tool Exposure

Expose only the tool capability groups relevant to the current Agent step where safely possible, with conservative expansion and a full-capability fallback when needed.

Final Walkthrough

Produce a grounded work-result artifact that distinguishes what was implemented, validated, reviewed, inferred, skipped, or still unresolved.

Workspace Search & Replace

Search across the workspace, preview replacement plans, and apply controlled replacements while preserving editor and workspace safety boundaries.

Governed Web Search & Fetch

AOH-owned web search and bounded web fetch are available under Network & Send Data approval policy, separate from HyperRouter's model-routing responsibility.

A Complete Agentic Development Workflow

The Release Candidate gate validates an end-to-end workflow rather than isolated chat features.

Understand Search Plan Edit Build Diagnose Fix Rebuild Test Review Walkthrough
Human approval remains authoritative. A plan can be reviewed before implementation, tracked source changes remain subject to AI Diff Review, and validation results are reported from actual build, test, runtime, diagnostics, and review evidence.

One Workbench. Multi-Provider Intelligence.

AOH does not expose alternative direct model backends. HyperRouter is the fixed AI connectivity layer, while remaining a separate product that can also be used by compatible IDEs and coding clients.

Auto RollOver

When an eligible upstream attempt fails before downstream output is committed, HyperRouter can continue through its configured provider routing and fallback policy without requiring AOH to switch backend configuration.

Local & Cloud AI

HyperRouter can provide AOH with Local Device GGUF, user-owned Local Cloud, Direct BYOK cloud providers, and its configured fallback cascade through one stable AOH backend identity.

Agentic Continuity Backend

HyperRouter normalizes supported protocol and tool traffic and maintains portable canonical history so compatible Agent workflows can retain usable context across eligible provider changes.

Provider Configuration Stays in HyperRouter

API keys, physical provider/model selection, routing priority, Local AI configuration, quota handling, and provider-specific compatibility remain owned by HyperRouter rather than duplicated inside AOH.

Explore AlphaOne HyperRouter

Review AI Work Before You Accept It

AlphaOne Harness is designed around governed Agent execution rather than silent source mutation. Implementation plans can be reviewed, sensitive tool operations remain subject to policy, tracked Agent changes enter AI Diff Review, and the user retains explicit Accept/Reject authority.

Plan Review & Proceed
Approval & Permissions
File & Hunk Review
Explicit Accept / Reject
Checkpoint Restoration
Evidence-Based Walkthrough

What Comes After the Release Candidate

The following phases are planned post-RC capability expansions. They are not presented as current Release Candidate features.

PHASE 10

Universal Content Reading & Document Intelligence

Extend bounded reading and provenance-aware extraction to common engineering and business document formats without expanding write authority.

POST-RC / PLANNED
PHASE 11

Workspace Knowledge / RAG

Add persistent, incremental workspace retrieval for large repositories and document-heavy workspaces while keeping current source and session evidence authoritative.

POST-RC / PLANNED
PHASE 12

Full Source Control / Git

Expand the current Git foundation with richer changes/diff, stage/unstage, commit, branch, history, stash, pull/push, and AI-assisted source-control workflows.

POST-RC / PLANNED
PHASE 13

LSP / Code Intelligence

Add Language Server Protocol intelligence including diagnostics, symbols, definitions, references, rename, completion, hover, and code actions.

POST-RC / PLANNED
PHASE 14

Advanced Agent Platform

Expand Agent Profiles, tool sets, permission profiles, Agent handoff, Skills management, and MCP management using the existing runtime where appropriate.

POST-RC / PLANNED
PHASE 15

Tasks, Test Explorer & Debugging

Add richer task/run workflows, Test Explorer, debugger foundation, breakpoints, watch, call-stack, and related execution tooling.

POST-RC / PLANNED

AOH Setup Handles the HyperRouter Prerequisite

AlphaOne Harness and HyperRouter remain separate installed products. AOH Setup checks the installed HyperRouter version and installs or upgrades the bundled compatible prerequisite when required, while preserving the same or newer compatible installation.

HyperRouter not installed AOH Setup installs the bundled compatible HyperRouter prerequisite.
Older HyperRouter installed Setup upgrades it to the compatible bundled version required by the AOH package.
Same or newer compatible version The existing HyperRouter installation is preserved; AOH does not downgrade it.
Product provenance: AlphaOne Harness is an independent AI coding workbench originally derived from DeepSeek Harness (DSH), with substantial AlphaOne-specific architecture, capabilities, security controls, and workbench enhancements. Third-party license and attribution notices are shipped with the product.