01 — ORCHESTRATE
Continuous task flow
Break product goals into bounded work, dispatch specialized agents, ingest their reports, and keep the next step moving without relying on manual copy-paste coordination.
AI-native Software & SaaS · Vietnam
Engineering orchestration for AI agent teams.
HaTrung AI is developing a supervisory layer that turns product goals into structured work, coordinates specialized coding, review, testing, and audit agents, and keeps parallel execution aligned with the project's technical source of truth.
Bootstrapped
Founded 2025
Prototype in active development
Human-in-the-loop by design
01 / What we are building
Make multi-agent development dependable, not just impressive.
The hard part is no longer getting an AI agent to write code. The hard part is keeping many agents coordinated over a long-running project without context drift, overlapping work, broken handoffs, or silent divergence from architecture.
02 — ISOLATE
Parallel work without collisions
Use isolated workspaces, explicit ownership, technical source-of-truth documents, and merge gates so workers can move in parallel without overwriting one another.
03 — VERIFY
Review before progress
Route outputs through reviewer, test, and audit stages. Important decisions stay visible to humans while routine coordination continues automatically.
A long-running control loop
Designed for real software projects where agents must wake up, hand off state, recover from failure, and continue until an approved outcome is reached.
Product goalRequirements, constraints, architecture, acceptance criteria
SupervisorPlan waves, route work, preserve project state
Specialist agentsCoding, research, testing, debugging in isolated workspaces
Quality gatesReview, audit, tests, conflict and drift checks
Approved changeHuman-visible decision and continuation to the next task
02 / Claude integration
Claude as the reasoning and supervisory layer.
Our integration focus is using Claude where high-quality judgment matters most: planning, technical decisions, long-context project reasoning, code review, agent supervision, and recovery when a workflow leaves the happy path.
Integration focus
From prototype to a reliable Claude-centered production loop.
We are actively designing the product around model routing, durable context, observable tool use, evaluation, and human approval boundaries rather than treating the model as a one-shot chatbot.
Claude APIReasoning, supervisory decisions, structured handoffs, and model routing.
Claude CodeEngineering workflows where implementation context and tool execution must stay close together.
MCP & tool useControlled access to project state, repositories, reports, and task-specific actions.
Evals & observabilityMeasure handoff quality, review reliability, recovery behavior, latency, and cost.
03 / Current stage
Early-stage, focused on reliability before scale.
The platform is in active prototype development. Current work prioritizes orchestration continuity, worktree isolation, agent wake-up and handoff mechanics, context discipline, auditability, and cost/latency-aware model selection.
Product principle
Autonomy with boundaries.
Agents should be able to continue routine work without constant human prompting, while high-impact decisions remain inspectable and interruptible.
- NowHarden the orchestration loop and failure recovery.
- NextDeepen Claude integration, evals, telemetry, and routing.
- ThenValidate with small engineering teams before broader release.
What we want to learn
Production patterns that hold up.
We are especially interested in guidance around long-running agent reliability, context management, MCP and tool design, evaluation methodology, human-in-the-loop controls, and scaling Claude workloads without sacrificing reasoning quality.
- QualityKnow when a supervisor decision is actually good.
- StatePreserve useful context without carrying unnecessary history.
- ScaleRoute the right model to the right engineering decision.
Company
HaTrung AI
A bootstrapped Vietnam-based Software & SaaS project building AI-native infrastructure for coordinated software engineering agents.
FoundedFebruary 2025
StagePrototype / early-stage
IndustrySoftware & SaaS
BaseVietnam
Build with us
Reliable AI agents need more than a prompt.
Privacy
This public landing page does not intentionally collect personal information. If you contact HaTrung AI by email, the information you provide is used to respond to your inquiry. Hosting and security providers may process standard technical logs needed to operate and protect the site. Internal operational tools are access-controlled and are not part of the public product experience.
Terms
HaTrung AI is a product under active development. Descriptions on this website communicate current product direction and may evolve. Nothing on this page constitutes a guarantee of feature availability, service level, pricing, or commercial terms.