#77th of 23 in Agent backends
frontieragent
Terminal agent runtime with single-agent and parallel sub-agent research workflows
- Stars
- 5.3k
- License
- Apache-2.0
- Last commit
- Oct 2026
- Language
- Python
7 on 19 Aug 20265,262 on 14 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
FrontierAgent is a Python terminal UI and agent runtime for long-horizon research and file-based work. It has two modes: a stateful ReAct agent, and an Agent Team where a coordinator delegates to parallel sub-agents through a task board. Shell and file tools run in a task-scoped sandbox with /inputs, /workspace and /outputs, with diff approval, checkpoints and /revert. It also bundles a benchmark runner for BrowseComp, HLE, GDPval and others.
Who it is for: Engineers running long-horizon research agents or benchmarking agent workflows
Strengths
- Agent Team mode with live task board and bounded parallel sub-agents
- Read-only /inputs, writable /workspace and /outputs; mutations need approval and are revertable
- Works with any OpenAI-compatible endpoint; optional keyless Parallel MCP for web search
- Bundled benchmark harness with resumable runs and per-failure reruns
Weaknesses
- Aimed at research and file deliverables rather than code editing in an IDE
- Agent Team concurrency multiplies with runner concurrency, raising simultaneous model calls
- Prebuilt GHCR image is private and needs docker login; otherwise build from source
- Local SGLang serving depends on pinned NVIDIA driver and CUDA tracks
What it needs
- Python, OpenAI-compatible endpoints, Apodex-1.1, SGLang-served models
- Needs Python 3.12, uv, OpenAI-compatible model endpoint, Serper or Jina API keys (optional), Docker (optional), SGLang (optional)
- Docker
- env example file
Also in Agent backends
See all 23| Rank | Project | Score |
|---|---|---|
| 1 | deepseek-harnessPlugin-based agent harness from DeepSeek with a local Web UI | 70 out of 100 |
| 2 | eve-software-factory-templateeve pipeline that turns GitHub or Linear issues into reviewed draft PRs | 59 out of 100 |
| 3 | computerDurable Object virtual filesystem with container, shell and JavaScript runtimes for agents | 58 out of 100 |
| #4 | marketing-team-eve-templateeve lead agent delegating to five marketing specialists with approval gates | 55 out of 100 |
| #5 | adk-recipesRunnable Agent Development Kit recipes, from single patterns to deployable agents | 53 out of 100 |