The operating framework for accountable AI engineering
Engineering work, made accountable.
Approved intent. Visible work. Human judgment.
Bring an app idea. Keep direction and delivery in view.
Basar AI Engineering OS is a digital software organization under development. In the planned experience, you describe an app idea to Hermes and clarify a bounded brief: intended users, outcome, constraints and budget. Basar keeps the authoritative project record while Hermes coordinates specialist execution autonomously within approved boundaries. You review the proposed plan, a versioned software candidate and the evidence behind decisions that need your authority.
Product direction — planned operating model; availability depends on the verified release.
For founders, teams and software agencies, the intended value is less review effort, less avoidable rework and clearer handoffs, while retaining direction over the outcome.
Product direction — planned operating model; availability depends on the verified release.
01Planned capability
AI organization
Illustrative model
Product and design
Orchestration and build
Independent review
Hermes assembles temporary teams around each outcome and coordinates their work within approved boundaries. People retain authority over material scope, architecture, risk, budget and release decisions.
02Planned capability
Governed workflows
Illustrative model
Approved scope
Review and approval gates
Recovery and revision
Build from an independently challenged architecture and a human-approved delivery plan. Planned controls would enforce scope, permissions and resource limits, with recovery, revision and exceptions kept accountable.
03Planned capability
Shared product truth
Illustrative model
Requirements and context
Decisions and rationale
Artifacts and history
Keep versioned requirements, approvals and evidence in the project record. Give each role relevant, bounded context. Retain validated lessons after review and approval; raw chat and worklogs do not become policy.
04Planned capability
Evidence and economics
Illustrative model
Tests and review findings
Provenance and outcomes
Usage and cost attribution
Connect accepted outcomes to tests, reviews and artifacts. Attribute usage, cost and budget impact to tasks and projects where verified. Measure cost per accepted outcome; token counts alone are not success.
Product direction — planned operating model; availability depends on the verified release.
Runtime enforcement of the full blueprint-assurance gate remains in development. These commitments describe the intended operating model.
The planned delivery loop02
Every change, understandable. Every step, accountable.
The planned journey starts with an app idea and a bounded brief developed with Hermes. Approved requirements, an independently challenged blueprint and a delivery plan guide the work toward a candidate you can review.
01Human-owned
Approved intent
Illustrative model
Desired outcome
Scope and constraints
Acceptance criteria
Clarify the app idea with Hermes through discovery: intended users, outcome, scope, constraints, budget and acceptance criteria. The founder approves the requirements; an initial prompt does not authorize implementation.
02Before build
Assured architecture
Illustrative model
Fixed architecture proposal
Independent challenge
Approved delivery plan
Challenge a fixed blueprint independently of its author. Resolve blocking findings, then obtain human approval of the architecture and delivery plan before implementation starts.
03Planned execution
Bounded implementation
Illustrative model
Specialist responsibilities
Work and change records
Continuous security and QA
Hermes coordinates tasks, dependencies, sequencing and recovery within the approved blueprint and plan. Specialists build with continuous security and QA; the founder steers outcomes without managing every agent task.
04Evidence required
Evidence and review
Illustrative model
Tests and artifacts
Independent review
Corrections and retests
Consolidate continuous testing and independent review into current, versioned evidence. Check acceptance criteria, connect findings to corrections and retests, and surface unresolved risks before candidate review.
05Accountable owner
Human decision
Illustrative model
Versioned candidate
Evidence and known risks
Revise, hold or release
Review a versioned candidate with a preview where supported, release notes, evidence and a recovery plan. Request revisions or hold; production release requires explicit human authorization for that candidate.
In the planned lifecycle, preview feedback, operational signals and analytics lead to scoped revisions and new candidates. Lessons are reviewed before becoming shared knowledge; none of these signals silently changes approved scope. Preview availability depends on the verified release.
Product direction — planned operating model; availability depends on the verified release.
Learn / Role constellation03
Specialized work. Shared direction.
Clear responsibilities, coordinated around approved intent.
Shared context
Approved scope Evidence and decisions
Strategy / Product
Clarify founder intent; propose scope and acceptance criteria.
Design
Shape usable, accessible experiences.
Architecture
Prepare the blueprint for independent challenge.
Engineering
Build from the approved blueprint and plan.
Quality / Security
Test continuously and surface security risks.
Operations
Plan reliable delivery; monitor operations and recover safely.
Illustrative role map - not an active agent roster.
Temporary teams are assembled around each outcome. Each role receives only relevant, bounded context. Raw chat and worklogs do not become authoritative memory without review and approval.
The planned operating model
How the work connects.
Basar
The planned authoritative project record: scope, approvals, assignments, worklogs, evidence, usage and delivery status.
Hermes
Your planned conversational orchestrator: clarify the brief, coordinate specialists and report progress within approved boundaries. Hermes cannot grant approval.
Human authority
People approve direction, architecture, material risk, budget, release and exceptions. Agents execute within those decisions.
Security and quality are intended to remain active throughout delivery.
Product direction - planned operating model; availability depends on the verified release.
Decide / Accountability trail04
The outcome matters. So does the record.
Basar is designed to retain the evidence behind delivery across tasks, sessions and handoffs. A reviewer should be able to establish what was authorized, what changed, what supports completion and what remains unresolved.
Product direction — planned operating model; availability depends on the verified release.
Conceptual illustration of governed delivery, not a product screen.
Work & provenance
Assigned role, objective, task state, start/end boundaries and worklog history
Artifacts & outcomes
Concrete artifacts linked to substantiated accomplishments and accepted outcomes
Verification
Current test and independent-review evidence tied to the version being assessed
Human decisions
Scoped approvals and decisions bound to the relevant artifact or candidate version
Continuity
Audit events, rationale and change history, with validated lessons separately reviewed and promoted
Resource attribution
Model/provider identity, input/output tokens and task/project cost and budget attribution where verified
Illustrative accountability trail — not live project, audit, or telemetry data.
Product direction — planned operating model; availability depends on the verified release.
Learn / Operating visibility05
Visibility that supports judgment.
The planned operating view puts project health, delivery phase, outcome progress, budget, aggregate usage and pending decisions first. Drill into agent tasks for their objectives, states, worklogs, evidence and attributable usage.
Task state
Distinguish planned, active, blocked, completed, failed or cancelled work. Track pending human approvals separately.
Evidence
Inspect the versioned artifacts, tests and reviews that support a claim.
Accomplishment
Distinguish activity from a substantiated outcome.
Usage visibility
Inspect task and project consumption, model/provider attribution where available, and budget impact.
Planned visibility — no live task, usage, provider, or cost telemetry is shown.
Product direction — planned operating model; availability depends on the verified release.
The test is a real delivery.
A credible system needs demonstrated controls and accepted outcomes. Repeated deliveries should show practical value for founders and teams.
Validation goals, not demonstrated outcomes. This page presents no completed delivery or customer-validation evidence.
Trace an actual outcome
Follow one real app from approved requirements through implementation, audits and corrections to a versioned candidate, acceptance evidence and a recorded human release decision.
Demonstrate the controls
Show that a missing approval or failed required check prevents the relevant consequential action.
Establish repeat value
Measure review effort, avoidable rework and cost per accepted outcome across repeated deliveries. Check that founders and teams can use the record to make better decisions.