Skip to content
AKZ

Personal R&D · AI Agents / Architecture

AI-HQ

A personal engineering platform designed to coordinate specialized AI agents around real software projects while keeping developers in control of execution, context, knowledge and quality.

Status
In development
Source
Private source
Period
2026 — ongoing
Role
Sole designer and engineer
AI Agents & Multi-Agent SystemsSoftware Architecture & System DesignFull-Stack Software DevelopmentTypeScriptNode.jsProvider abstractionsAutomated test & audit gatesLocal LLMs (Ollama)

01Interactive exhibit

Show

  1. Developer control

    01

    Design intent: the developer stays in control of execution, context, knowledge and quality.

  2. Provider substrate

    02

    Phase 7, independently audited. Reported validation: 1,250 tests with zero failures across 34 typecheck projects (per the project's own audit notes).

  3. Orchestration

    03

    Roadmap. Not published as completed features.

Provider substrate

Billing mode

Validated

Billing is explicit: an available API or subscription must never silently imply permission to incur usage-based charges.

Select a node to read what it does and how far it has been built.

Conceptual diagram · not live telemetry

02

Context & challenge

Using several AI tools can improve productivity, but unmanaged agent collaboration also creates duplication, context loss, unpredictable cost and difficult-to-verify results.

I wanted a platform where different specialized agents could contribute as one engineering team without making the workflow opaque or dependent on a single provider.

03

Approach

AI-HQ is designed as a provider-agnostic command layer rather than a collection of unrelated chatbots. Projects receive profiles appropriate to their needs (frontend, backend, full-stack, mobile or custom) and the system is designed to activate the smallest useful specialist team.

The architecture explicitly models providers, provider connections, runtime/transport, authentication method, billing mode, models, capabilities and agents as distinct concepts. An available API or subscription must never silently imply permission to incur usage-based charges.

The wider design also includes reusable engineering knowledge, dependency-aware execution plans, controlled parallelism, reversible changes, transparent handoffs and meaningful engineering benchmarks. These are architectural goals and staged implementations, not a claim that every capability is already complete.

04

My role

Personal R&D project: I own the product direction, architecture, implementation and validation strategy.

05

Engineering decisions

Minimum necessary coordination

Use fewer agents when fewer suffice.

Explicit execution boundaries

Keep permissions, provider credentials and billing behaviour visible and controlled.

Measurable quality

Optimise for trustworthy results, not the number of agents or generated lines.

Reversible evolution

Support rollback, auditing and knowledge preservation across architecture changes.

06

Validated

  • Hardened the foundation and validated successive platform phases using automated test and audit gates.
  • Completed an independently audited Phase 7 provider substrate, separating providers, runtimes, authentication, billing, models and agent capabilities.
  • Introduced explicit safeguards against unapproved billing-boundary fallback, inferred credentials or silent secret persistence.
  • Phase 7 reported validation suite: 1,250 tests with zero failures across 34 typecheck projects, plus related build and plugin checks (as documented in the project's own audit notes).

07

Roadmap — not yet completed

  • TeamLead-driven task decomposition into dependency-aware plans and checkpoints.
  • Specialist roles such as Codebase, Code, UI, QA/Test, Performance, Research and Agent Builder.
  • Knowledge retrieval that favours relevant deltas, reuse and source-backed decisions over repeated full-context transmission.
  • Scheduled research and maintenance orders with change detection and concise reporting.
  • Evaluation suites measuring correctness, latency, token use, execution cost, retries, handoffs, unnecessary edits and domain outcomes.
  • A unified local desktop/terminal/sidebar workflow and future native provider adapters where appropriate.

08

Status & outcome

A validated foundation for an ambitious multi-agent engineering platform, with the main orchestration workflows still under development. Demanding end-to-end benchmark projects are intentionally reserved for later roadmap stages.

09

Attribution & publication notes

  • The diagram on this page is conceptual. It shows the architecture's concepts, not live execution telemetry or screenshots of the application.
  • Repository is private.

Next project

UTEST Industrial Testing Platform