iGent

Maestro clones repositories, installs dependencies and runs shell commands inside secure cloud sandboxes, supports Python and JavaScript out of the box with the vendor describing it as extensible to further languages and frameworks, handles projects of thousands of files across multiple languages, and can expose a sandbox web server externally for sharing a running result. The vendor states it integrates the tools and services developers already use, but names none: no source control platform integration beyond repository cloning, no ticketing, chat, CI or MCP connection is documented on igent.ai or the insights posts.

Sourceigent.ai home and signup pagesread 2026-09-29

Maestro works across the development lifecycle from a prompt, an uploaded codebase or a cloned repository, planning, implementing, executing, testing and validating in a cloud sandbox with adaptive autonomy shifting between interactive collaboration and independent execution over sustained sessions; the vendor demonstrates Ferrous, a Redis-compatible database of roughly 35,000 lines written entirely by the agent, and reports drafting solutions to all twelve ICPC 2025 World Finals problems in about two hours without human involvement. No multi-agent coordination, subagent delegation or documented handoff between agents appears on igent.ai or the insights posts; the ensemble described is of models rather than agents.

Sourceigent.ai/insights/production-database-server and igent.airead 2026-08-30

Maestro handles projects with thousands of files across multiple programming languages, understanding structures and patterns across a codebase and adapting to a project's own patterns and style with no fine-tuning required, working from a prompt, an uploaded codebase or a cloned repository; the vendor attributes this to advanced in-context learning and inference-time reasoning, and has publicly credited million-token context models for building large systems. No codebase index, retrieval layer, dependency graph, semantic search or document grounding surface is documented.

Sourceigent.ai and igent.ai/insights/introducing-maestroread 2026-09-29

Maestro offers adaptive autonomy, shifting between tight collaboration and independent problem solving so the developer sets how much runs unattended, from interactive brainstorming through to fully autonomous execution, and the vendor states the developer sets goals and retains control while providing strategic direction and approval; because work runs interactively the developer observes and redirects as it proceeds. No approval gate, permission scoping, policy configuration, spend or step limit or other shipped guardrail mechanism is documented on igent.ai or the insights posts.

Sourceigent.ai home and signup pagesread 2026-09-29

Code runs in secure cloud-based sandboxes with the vendor stating that code stays private and secure, and sandbox isolation means execution happens in a contained environment rather than on shared infrastructure; a sandbox web server can be deliberately exposed externally for sharing, which is a customer-initiated action. No security certification, attestation, trust center, single sign-on, role-based access control, audit logging, data processing agreement or security page is documented on igent.ai or the insights posts, despite the vendor targeting enterprise engineering teams.

Sourceigent.ai home and signup pagesread 2026-09-29

Developers watch Maestro generate, test and refine code in real time within the sandbox, giving direct visibility into its actions as they happen, and the sandbox can expose a running web server so results can be inspected and shared; test, lint and type-check output is visible as the agent works. No retained execution trace, session history, replay, audit log or cost reporting surviving the session is documented on igent.ai or the insights posts.

Sourceigent.ai home and signup pagesread 2026-09-29

A learning claim with no store behind it. codemaestro.ai, iGent's own earlier product page, says the assistant learns the user's style, preferences and common tasks with each conversation, and igent.ai says Maestro adapts to a codebase's patterns without fine tuning. Neither page names a memory store, its scope, lifetime or a way to review or delete it.

Sourcecodemaestro.ai and igent.airead 2026-09-29

Maestro runs exclusively in the vendor's secure cloud-based sandboxes, with projects started from a prompt, an uploaded codebase or a cloned repository and results shared by exposing a sandbox web server. No self-hosted, on-premises, VPC, private cloud, air-gapped or customer-selectable region option is documented, and no data residency statement, sub-processor list or data location commitment appears on igent.ai or the insights posts.

Sourceigent.ai home and signup pagesread 2026-09-29

One agent, Maestro, ships, and projects begin from scratch, from an uploaded codebase or from a cloned repository, with the agent adapting to whatever patterns it finds rather than applying supplied templates. No library of prebuilt agents, template gallery, starter pack, skills catalog or marketplace is documented on igent.ai, the insights posts or the vendor's GitHub organization, whose public repositories are demonstrations and forks rather than distributable assets.

Sourceigent.ai and github.com/iGentAIread 2026-09-29

Maestro is an interactive conversational development tool the developer drives directly, starting work from a prompt, an uploaded codebase or a cloned repository within a single web surface. No event-driven trigger such as a commit, pull request or issue, no scheduled or recurring run, no inbound chat, ticketing or source control channel, and no editor, terminal or command line entry point is documented on igent.ai or the insights posts.

Sourceigent.ai home and signup pagesread 2026-09-29

Maestro conducts an ensemble of frontier AI models from leading companies, selecting models per task, and the vendor has publicly named Claude Sonnet 4 with one million tokens of context as the model used to build its Redis-compatible database demonstration. Model selection is made by the vendor as part of the product rather than exposed to the customer: no model picker, routing configuration, bring-your-own-key path, custom endpoint or local model option is documented on igent.ai or the insights posts.

Sourceigent.ai and x.com/iGent_AIread 2026-09-29

The vendor describes Maestro as open and extensible to additional languages and frameworks and as integrating the tools and services developers already use, and the sandbox executes arbitrary shell commands, which is a generic extension path.

No public REST API, SDK, MCP server, webhook surface or plugin interface for the Maestro platform is documented on igent.ai or the insights posts, there is no documentation site, and access to the product itself is through a waitlist with incremental rollout.

The vendor's GitHub organization includes a fork of the litellm gateway, which is infrastructure it consumes rather than a surface it exposes.

Sourceigent.ai and github.com/iGentAIread 2026-09-29

Maestro executes, tests and validates code in secure cloud sandboxes with built-in linting, typing and style checks, and can benchmark and optimize performance, which the vendor frames as grounded engineering that reduces errors; the demonstration database is reported as outperforming established alternatives on core operations.

These verify the agent's own output rather than providing a harness the customer can point at their own workload, and no evaluation product, benchmark suite or agent regression testing surface is exposed to customers. The vendor's GitHub organization carries a fork of the SWE-Lancer benchmark and a repository of its ICPC 2025 solutions, both vendor-side benchmarking.

Sourceigent.ai, igent.ai/insights/production-database-server and github.com/iGentAIread 2026-08-30

Maestro operates entirely through programmatic interfaces: the sandbox filesystem, shell command execution, dependency installation, repository cloning and test and lint tooling. It can make a sandbox web server externally accessible so a human can view or demonstrate the running result, which is publishing an endpoint for a person rather than the agent operating a rendered interface. No browser control, screenshot capture, visual verification or operation of software lacking a programmatic interface is documented on igent.ai or the insights posts.

Sourceigent.ai home and signup pagesread 2026-09-29