Amar Studio
The development agency piloted by persistent state.
A state-piloted development platform that integrates the discipline of software engineering into professional workflows — for users who are not developers but need the rigor of a dev team.
AI dev tools are conversational and lose the thread.
Today's AI coding assistants are chat-based: they work within a conversation, and when the conversation ends, the context is gone. Resuming a project means re-explaining everything. And for professionals who are not developers — finance, accounting, operations teams — there is no tool that brings the rigor of a development agency into their own workflows.
Amar Studio runs an entire software project lifecycle driven by persistent, versioned project state rather than chat history. A new agent reads the state and continues — no conversation needed.
A multi-agent runtime, governed by versioned skills.
Amar Studio is genuinely multi-provider and multi-model: Anthropic Claude, Google Gemini, OpenAI GPT, xAI Grok, Ollama, and Google Antigravity all feed a shared agent runtime that reads state, executes eligible work with versioned skills, and publishes proofs.
State-piloted project lifecycle
The canonical cycle — discovery → specs → solution → planning → build → quality → delivery → learning — is driven by a persistent, versioned project state, not by chat history. Resumption works without re-explaining anything.
Versioned skills & roles
Method is separated from the agent that applies it. Skills evolve via reviewed amendments — never silent overwrites. A dedicated Agent Librarian is the sole authority to modify skills.
Evidence-driven completion
No 'completed' status without criteria and proofs: SHA-256 file proofs, test results, quality gates. An impediment keeps work blocked with an owner and required action — no silent failures.
Semantic skill recommender
A local LLM (Ollama) selects which versioned skills to load for a given task — the runtime assembles only the context the agent needs, rather than dumping the whole system into the prompt.
State-first. Role-first. Evidence-driven. Clean as you go. Governed feedback. The value is the ability to drive a project to a verified, explainable, resumable, and governable outcome.
Everything in one place
Persistent, versioned project state
Objectives, work items, artifacts, decisions, proofs, risks, debt, feedback, and an append-only event journal with derived snapshot.
Multi-role runtime with handoffs
PO, Analyst, PM, Designer, Architect, Tech Lead, Dev, DevOps, Reviewer, QA, Writer, Librarian — parallel orchestration with Git worktrees and locks.
MCP server integration
Exposes the tool registry as an MCP server for external agent integration — filesystem, terminal, Git, browser, Forgejo, build/QA.
13+ workbench views
Portfolio, Project Cockpit, Specs, UI Design, Planning, Execution, Issues, Quality, Documentation, Tests, CI/CD, Artifacts, Agency, Skills — each role-specialized.
Finance & accounting vertical
A real use-case pilot: payroll verification, fraud detection, bank reconciliation, payment matching, cash-flow categorization, audit trail, monthly close validation.
Self-improvement loop
Micro and macro feedback → amendments → librarian review → new tested skill versions — the system learns and improves over time.
Built on a modern, proven stack
Desktop app
Backend
Storage
AI providers
Tooling
Quality
Where we stand today
Request early access
Get in touch to request early access or learn more about this product.