iFixAi
ifixai-ai/iFixAiReviewed by AI
What it does
A command-line tool for people who build AI agents. It puts your agent in a simulated version of your business, sends the kind of requests a real day brings, including requests that push the agent past what it is allowed to do, and another AI model grades each reply. You get a pass or fail per check, an A-to-F grade and reports in JSON and Markdown. You need Python 3.10 or newer and a paid model key for the judge. A demo mode with canned answers runs with no key. It also installs as a plugin or skill inside Claude Code, Codex, Cursor and similar.
How easy it is to install
C · install 1 or 2 tools · Windows
- Windows C
- macOS C
- Linux C
There are no ready installers in the 23 releases, so on every platform you install Python 3.10 or newer, run `pip install ifixai` and then a few commands; Windows, Mac and Linux are all C. The README adds Windows-specific steps (`py -m venv`, `python -m ifixai.cli.main`) and warns that a stock Mac ships Python 3.9, which is too old.
Numbers on GitHub
- Stars
- 41,458on Oct 31, 2026
- Downloads
- —all releases
- Growth
- +6,394stars in 5 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Python
- To build from source
- Python
- Disk space
- 100 MB–300 MB · estimate
- Accounts or keys
- OpenAI, Anthropic, Google Gemini, Azure, Bedrock, OpenRouter or Hugging Face (any one can be the judge instead of OpenAI)
- Cost
- Free, API costs extra · The code is free (Apache-2.0). A real audit calls a paid model as the judge, plus your own agent's API; the README says `--dry-run` shows how many judge calls a run will make before anything is billed.
- License
- Apache-2.0 (Permissive)
- Can replace
- LangSmith, Braintrust
- Last update
- Oct 30, 2026
Risk signals (2)
- Attentionsends adversarial requests to your agent
README: "Use a staging copy, because the probes are real requests." The default run includes probes for prompt injection (`B12`) and for destructive actions by someone not allowed to ask for them (`P01`), so it should only be pointed at an agent you are allowed to test.
AI review
- Infovery young repository with large star count
The repository was created on 2026-04-27 and already shows 37,084 stars with only 3 open issues; the README opens with Product Hunt and Trendshift promotion badges.
AI review
Evidence is shown in its original language.
What Radar could not check
- Whether the audit grades are accurate, or whether a pass here means your agent behaves well outside the simulated scenarios.
- What a real audit costs in judge tokens, especially the `--full-run` with all 61 inspections.
- Whether the Claude Code plugin, the Codex plugin and the editor skills install and work as described.
- Whether the install works on each platform and which Python versions actually run it.
- What the tool sends to the judge provider, and whether any of your agent's answers or business data leave your machine during a run.