PLATFORM VISION

The agentic orchestration of software development

Airchor is evolving from a live planning and assessment workspace into a governed control plane for coordinating project intelligence, decisions, permissions, agents, tools, execution, verification, and delivery.

The architecture described on this page is the planned product direction. It is not current MVP v1.0 functionality.

PLATFORM VISION

The Airchor orchestration field

  1. Project context
  2. Airchor control layer
  3. AgentTasks
  4. Execution
  5. Verification
  6. Delivery
Project intelligence passes through a governed control layer before bounded execution, independent verification, and reviewable delivery.
Read visual as text
  1. Project context
  2. Airchor control layer
  3. AgentTasks
  4. Execution
  5. Verification
  6. Delivery

Context → control layer → AgentTasks → execution → verification → delivery

WHY ORCHESTRATION MATTERS

Coding is accelerating. Control is not.

AI tools can generate and modify software quickly, but the surrounding project state often remains fragmented: requirements, evidence, architecture, permissions, risk, tests, approvals, and delivery records live in separate places or disappear between prompts.

Agentic software development needs more than an autonomous model. It needs coordinated context, bounded authority, visible execution, and independent verification.

Isolated automation

Prompts, agents, repositories, CI, deployment, evidence, and approvals operate in disconnected systems.

Governed orchestration

A control plane coordinates context, bounded authority, visible execution, independent verification, and durable project state.

DEFINITION

What orchestration means

Orchestration is the coordinated management of tasks, systems, tools, permissions, and processes so they work together in the correct sequence toward a defined outcome.

For Airchor, orchestration means coordinating project context, requirements, decisions, evidence, repositories, agents, capabilities, approvals, execution environments, tests, builds, pull requests, previews, verification, and updated project state.

PLANNED CONTROL PLANE

Airchor as the control plane

Airchor should not replace coding agents, Git hosting, CI systems, deployment platforms, or developer tools. It should coordinate them while preserving intent, scope, permission, evidence, and the recommended next action.

PLANNED ARCHITECTURE

Nine control-plane layers

  1. Project intelligence
  2. Planning & decisions
  3. Agent orchestration
  4. Capabilities & permissions
  5. Execution
  6. Verification
  7. Evidence & audit
  8. Integration adapters
  9. User execution console
Nine planned layers separate project intelligence, planning, orchestration, permission, execution, verification, evidence, integrations, and the user console.
Read visual as text
  1. Project intelligence
  2. Planning & decisions
  3. Agent orchestration
  4. Capabilities & permissions
  5. Execution
  6. Verification
  7. Evidence & audit
  8. Integration adapters
  9. User execution console

A provider-neutral control plane coordinates specialist systems without replacing them.

LIVE FOUNDATION

The orchestration foundation already exists

MVP v1.0 already provides persistent projects, two route-specific workflows, evidence and source-of-truth records, readiness and risk indicators, roadmaps, builder and developer handoffs, generated documents, and saved project state.

NINE PLANNED LAYERS

The planned Airchor platform

Each layer has a distinct responsibility. Together they preserve context, authority, execution evidence, verification, and the updated project record.

PLANNED

Project Intelligence Layer

Repository facts, project context, evidence, source citations, findings, assumptions, and immutable snapshots.

PLANNED

Planning and Decision Layer

Scoped objectives, exclusions, acceptance criteria, dependencies, plan versions, and explicit stop conditions.

PLANNED

Agent Orchestration Layer

Durable AgentTasks, sessions, runs, event streams, pause, resume, retry, cancellation, and project-state updates.

PLANNED

Capability and Permission Layer

Default-deny capabilities, task-scoped grants, approval records, expiry, revocation, environment restrictions, and action digests.

PLANNED

Execution Layer

Registered runners or managed sandboxes, isolated branches and worktrees, tools, commands, file changes, and bounded network access.

PLANNED

Verification Layer

Independent checks against the approved plan, exact repository state, tests, builds, scope, diffs, and delivery artefacts.

PLANNED

Evidence and Audit Layer

Ordered events, tool calls, artefact manifests, usage, approvals, verification results, and immutable history.

PLANNED

Integration Adapter Layer

Provider-neutral connections to coding agents, GitHub, GitLab, CI, Vercel, database providers, and future tools.

PLANNED

User Execution Console

A project-attached experience showing what is happening, what the agent may do, what requires approval, what evidence exists, and what happens next.

FIRST COMPLETE WORKFLOW

The first complete agentic workflow

The recommended first end-to-end workflow extends Work on Existing Project from a reported finding into a verified, reviewable repair.

This workflow is planned. It is not available in MVP v1.0.

PLANNED WORKFLOW

Governed AgentTask lifecycle

  1. Finding
  2. AgentPlan
  3. Human approval
  4. Capability grant
  5. Execution
  6. Independent verification
A finding becomes a plan, passes an explicit human approval and capability gate, then moves through execution and independent verification.
Read visual as text
  1. Finding
  2. AgentPlan
  3. Human approval
  4. Capability grant
  5. Execution
  6. Independent verification

Finding → plan → approval → capability → execution → verification

PLATFORM VISION

What is live and what is planned

Current product capability and future orchestration remain visibly separate. Planned architecture is not presented as executable product functionality.

Current Airchor foundation and planned platform direction
LIVE IN MVP v1.0

Structured project state

  1. Planning and assessment workspace
  2. Create New Project
  3. Work on Existing Project
  4. Persistent projects and Saved Projects
  5. 14-document Project Blueprint
  6. 10-document Assessment & Recovery Plan
  7. Readiness, confidence, evidence, risk, assumptions, and next actions
  8. Authenticated Markdown downloads
PLANNED / VISION

Governed agentic execution

  1. Repository intelligence and immutable snapshots
  2. Grounded AI and source citations
  3. Approved AgentTasks and AgentPlans
  4. Capability and permission gates
  5. Local and managed execution environments
  6. Coding-agent adapters
  7. Independent verification
  8. Pull requests and previews
  9. Recurring monitoring
  10. Enterprise and multi-agent controls

This is the planned platform direction, not current MVP functionality.

VERIFICATION LOOP

Delivery updates the project—not just the code

The planned loop returns execution evidence, verification, artefacts, and the new repository state to the durable project context before the next action is selected.

PLANNED

Completion is a claim until it is verified

01 · AGENT REPORTAgent-reported completionCompletion claim
02 · INDEPENDENTIndependent checksPlan · diff · tests · build
03 · EVIDENCEDVerified resultUpdated project state
Agent-reported completion is checked independently against the approved plan, exact diff, tests, build, artefacts, and delivery state.
Read visual as text
  1. Approved plan
  2. Agent-reported completion
  3. Independent checks
  4. Verified result
  5. Updated project state

Agent report ≠ verified result. Evidence closes the loop.

MVP v1.0 · LIVE

Start with the live foundation

Airchor MVP v1.0 is available today. The roadmap shows how the planning foundation develops into governed agentic orchestration.