FuXi Terminal AI Coding Agent
Fast, self-contained AI coding agent in your terminal with cost-aware model routing.
- Source repo
- fuxicodex/Fuxi
- Stars
- ★ 3.2k
- Last updated
- 2d ago
- License
- NOASSERTION
- Primary language
- Python
- FA score
- 37/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer wants to fix failing tests in a project by instructing the agent in natural language, letting it edit code and run the test suite.
- Not a fit if
- Teams that need fully offline use with no account sign-up
- Teams that must audit or modify the source (proprietary, closed source)
- Buyers who need third-party benchmark scores to evaluate coding ability
- Source review
- 37/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
FuXi is a terminal-first AI coding agent built in Go, shipped as a single static binary with no runtime dependencies. It provides a Think → Act → Verify agentic loop and supports multiple providers including OpenAI-compatible, Gemini, Bedrock/Vertex, with cost-aware routing and automatic failover. It includes 50+ built-in tools for file operations, shell, ripgrep search, LSP diagnostics, Jupyter, browser use, and parallel sub-agents. Safety guardrails include an AST safety classifier and fine-grained permissions. Sessions and memory persist to disk, with checkpoints and automatic compaction. It is extensible as an MCP client with hooks, skills, plugins, and slash commands. Installation is via scripts, and a wide range of commands are available in the fuxi CLI. The product is proprietary; source code is not public.
FuXi reads and edits code files, executes shell commands via bash or PowerShell, uses ripgrep for search, fetches web pages with web fetch, uses LSP diagnostics, runs Jupyter notebooks, drives browsers through tools, manages background tasks, and launches parallel sub-agents. It interacts with OpenAI-compatible APIs through built-in tools and MCP servers. The agent loop includes thinking, acting, and verification steps, with cost-aware routing and automatic failover. It generates configuration via fuxi init, checks environment with fuxi doctor and fuxi verify, authenticates via fuxi login, and proxies binaries via fuxi proxy. Session transcripts persist to disk, allowing resume, rollback, or forking.
- A developer wants to fix failing tests in a project by instructing the agent in natural language, letting it edit code and run the test suite.
- An engineer prefers to use different LLM providers from a single interface and automatically route requests based on cost.
- A user with MCP servers wants to interact with their tools through an agent that supports MCP.
- Teams in security-sensitive environments require command classification and permission controls to manage autonomous operations.
- Users working on long sessions want to retain context across sessions, with checkpoints and automatic compaction.
- A team wants to integrate a terminal agent in CI environments, using
setup-tokenfor headless authentication.
How do you install or deploy this agent?
curl -fsSL https://fuxicode.com/install.sh | bash # macOS / Linux
irm https://fuxicode.com/install.ps1 | iex # Windows PowerShellVerify with fuxi --version and fuxi doctor.
How do you use this agent?
First configure a model: run fuxi wizard or edit ~/.fuxi/config.yaml. Launch the TUI with fuxi. Use slash commands like /model, /compact, /cost within a session. Run fuxi doctor for environment checks, fuxi verify to test connection, and fuxi update for updates. Full guide: docs/usage.md.
What are this agent's strengths and limitations?
- Offers 50+ built-in tools covering file ops, shell, search, LSP, browser, and Jupyter.
- Cost-aware routing and automatic failover leverage multiple providers efficiently.
- Safety guardrails like AST classifier and fine-grained permissions enhance secure autonomous operation.
- Single static binary with no runtime dependencies simplifies deployment.
- Proprietary license; source code unavailable, limiting modification and forking.
- No scores on third-party benchmarks like SWE-bench, requiring self-evaluation.
- Benchmarks are self-run task sets, not third-party validated, so treat cautiously.
- No explicit documentation on required API keys or specific provider setup steps; users must infer from config examples.
How does this agent compare with similar options?
FuXi positions itself as a provider-agnostic alternative to Claude Code, offering a Think → Act → Verify loop and cost routing.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| FuXi Terminal AI Coding Agent This agent | 37 · Major gaps | CLIFreemium | ★ 3.2k | 2d ago | Python | OpenAI API · Claude API |
| zot Coding Agent | 59 · Major gaps | CLIFree + model costs | ★ 351 | 1d ago | Go | OpenAI API · Claude API |
| senpi | 74 · Some gaps | CLIFree + model costs | ★ 488 | today | TypeScript | OpenAI API · Claude API |
| OpenClaude — Universal Terminal AI Coding Agent | 42 · Major gaps | CLIFree + model costs | ★ 34k | today | TypeScript | Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows permission modes (default/plan/bypassPermissions), --auto classifier-gated, --dangerously-skip-permissions warning, audit logging, AST safety classifier, rollback/checkpoints, update checksum verification. However, implementation details are missing; permission granularity, user confirmation flow, data flow transparency, sensitive data handling (e.g., key storage) are not detailed. Dependency security has no evidence (no dependency manifest or vulnerability scan). External effects (shell commands, network updates) are mentioned but without specific controls. Rollback has checkpoints but no detailed mechanism. Source attribution has copyright notice but no contributor guidelines.
Self-consistency: README description aligns with docs, but no code verification. Dependency availability: no dependency manifest or third-party library info, cannot assess. Failure messages: fuxi doctor and verify commands exist, but no specific error handling examples.
Audience clear (developers, terminal users), scenarios diverse (code editing, command running, tool driving). Capability boundaries have tool list and permission modes, but not explicit limits. Trigger precision: slash commands and CLI flags exist, but no detailed trigger conditions. Environment fit: supports multiple platforms (macOS/Linux/Windows), environment variable configuration.
Information architecture clear (README TOC, docs folder). Install notes detailed (curl, PowerShell, CMD). Naming stability: version numbers but no API stability promise. Examples and FAQ exist (docs/faq.md). Known limitations: only mentions benchmark is self-run small-scale, no other limitations. License: proprietary, but no standard license identifier. Versioning changelog exists (CHANGELOG.md). Maintenance responsibility: SECURITY.md and SUPPORT.md exist, but no explicit maintainers.
Output usability: TUI and CLI exist, but no specific output format examples. Marginal value: cost-aware routing, multi-provider support, but no independent verification. Cost-benefit: /cost and /usage commands exist, but no specific cost data.
Claim traceability: benchmark report and doc links exist, but no specific data. Cross-source corroboration: no third-party verification. Fact-inference separation: README distinguishes claims and benchmarks, but some claims lack evidence.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Repository contains only documentation and benchmarks; product source code is not published, so actual implementation cannot be verified.
- License is proprietary with no standard license identifier, which may affect compliance.
- No evidence of dependency security, supply chain risk cannot be assessed.
- Benchmark is self-run, not third-party independent verification, results may be biased.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
FuXi 0.1.8 Release Announcement Release Date: September 22, 2026 We are pleased to announce the release of FuXi 0.1.8, a major update following 0.1.7. This release introduces rate-limit banners and an official plugin marketplace mirror, and delivers systematic stabilization and refinement across session storage, the engine, permissions, Windows compatibility, and the user interface. New Features - Rate-limit banners: unified rate-limit data now feeds subscription banners; usage beyond the limit is silently rolled into additional usage with clear notification. …View on GitHub ↗ -
0.1.6 v0.1.6
Release notes excerpt
FuXi 0.1.6 Release Announcement Release Date: September 7, 2026 FuXi 0.1.6 is now available. This release focuses on core experience consistency and stability: unified session and message handling, hardened login flow, cross-platform compatibility improvements, and a number of stability fixes. New Features - Unified session & message handling: chat messages are consolidated into a single message model, removing inconsistencies between multiple handling paths; message parent-child chains are preserved more completely, and partially-compacted history survives session resume. …View on GitHub ↗ -
0.1.2 v0.1.2
Release notes excerpt
FuXi 0.1.2 Release Announcement Release Date: August 6, 2026 FuXi 0.1.2 is now available, with major improvements across multi-window coordination, image capabilities, and the plugin ecosystem. New Features - Multi-window coordination: switch seamlessly between concurrent sessions without interrupting your current work. - Image captioning: non-vision models can now generate image descriptions automatically, enabled by default. - Vision capability is automatically detected from known provider capabilities, with no manual configuration required. …View on GitHub ↗
FAQ
Is FuXi free to use?
Do I need a specific API key?
fuxi login to use FuXi-managed models.How does FuXi handle permissions and security?
What happens if a provider goes down?
What if my session context is full?
/compact command to compress conversation history, or rely on automatic compaction to save tokens.