Repolode
Submit
ApodexAI

ApodexAI/FrontierAgent

An open-source agent runtime and terminal app for long-horizon research and file-based work, with ReAct and parallel Agent Team modes.

View on GitHubWebsite
Stars
5,323
Forks
313
Open issues
5
Last commit
4 hours ago

What is it?

FrontierAgent is an agent runtime, terminal UI, and evaluation suite for multi-step research and file-based work. The frontier-agent TUI offers two workflows: a single stateful ReAct agent that researches, reads files, writes deliverables and runs commands, and an Agent Team mode where a coordinator keeps a task board, delegates independent work to parallel sub-agents, and synthesizes their reports.

Agent tools share a task-scoped sandbox with a read-only /inputs directory, a /workspace for working files, and a /outputs directory for deliverables. Mutating operations show a diff and ask for approval unless --yes is set, sessions are checkpointed and traced locally, and /revert undoes session file changes. Users can type follow-up instructions while a run is active, and they are queued for the next safe turn boundary.

The same workflow engine drives a subprocess benchmark runner, and the framework, tools, workflows, and evaluation layers are separate packages. It targets an OpenAI-compatible model endpoint, with optional Docker and local SGLang setups.

Key features

  • 01Two workflows: a stateful ReAct agent and a coordinator-led Agent Team with parallel sub-agents.
  • 02Live task board in the TUI sidebar showing pending, active, completed, blocked and cancelled tasks.
  • 03Task-scoped sandbox with read-only inputs, working files and persistent outputs.
  • 04Approval prompts with diffs for file writes and risky shell commands, plus /revert for session changes.
  • 05Checkpoint, local trace and --resume support for saved sessions.
  • 06Built-in subprocess benchmark runner with resumable runs and failure reruns.

Use cases

  • Running multi-source research tasks that produce a written report from web sources and local documents.
  • Analyzing a code repository or a set of documents from the terminal with an agent that asks before changing files.
  • Splitting a broad investigation across parallel sub-agents and collecting their findings.
  • Benchmarking agent workflows on research and file-grounded datasets.

Topics

#agent-orchestration#ai-agents#multi-agent#terminal-agent#tui#llm-evaluation#openai-compatible#agentic-ai#agentic-framework#harness

FAQ

What is FrontierAgent?
An open-source agent runtime and terminal app for long-horizon research and file-based work, with ReAct and parallel Agent Team modes. FrontierAgent is an agent runtime, terminal UI, and evaluation suite for multi-step research and file-based work. The frontier-agent TUI offers two workflows: a single stateful ReAct agent that researches, reads files, writes deliverables and runs commands, and an Agent Team mode where a coordinator keeps a task board, delegates independent work to parallel sub-agents, and synthesizes their reports.
Is FrontierAgent free to use? What license does it have?
FrontierAgent is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is FrontierAgent still maintained?
Actively maintained. The last commit to FrontierAgent was on Oct 18, 2026.
What can I use FrontierAgent for?
Running multi-source research tasks that produce a written report from web sources and local documents. Analyzing a code repository or a set of documents from the terminal with an agent that asks before changing files. Splitting a broad investigation across parallel sub-agents and collecting their findings. Benchmarking agent workflows on research and file-grounded datasets.
Where can I get FrontierAgent?
The source code and installation instructions are on GitHub: https://github.com/ApodexAI/FrontierAgent

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects