- Category
- AI Agents
- Rank
- No. 1818Tools index
Previous survey · No. 1826 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- Hmbown
- GitHub
- 4 stars
- Latest release
- v0.1.0
- Date
About
Multi-perspective reasoning harness that queries multiple LLM CLIs in parallel with different perspectives.
What it does
Midtry turns one question into a side-by-side panel of model answers. It discovers compatible command-line assistants already installed, gives each a distinct thinking style, runs them concurrently, and preserves successful or failed results separately for manual comparison.
Why it's ranked here
The core utility is small but credible: one command exposes disagreements that a single assistant can hide. Its configuration and failure reporting make experiments practical. However, the ambitious scoring and consensus material in the documentation is not implemented by the supplied runner.
What's good
Perspective prompts are concrete rather than cosmetic. Conservative, analytical, creative, and adversarial framing encourages different checks and solution shapes. Users can select providers, cap concurrency, set timeouts, customize prompts, filter allowed tools, preview output, and inspect partial failures instead of losing the whole run.
Tradeoffs
Every real run depends on separately installed and authenticated model tools. Completion time follows the slowest selected provider, while usage costs remain outside Midtry's accounting. Results receive no automatic synthesis, factual verification, retry, cache, token report, or budget control. Output quality still depends on the underlying models.
How to use it well
Use it when reviewing code, debugging, or debating architecture warrants independent approaches and manual judgment. Keep the provider set small when latency matters, then compare assumptions and edge cases across answers. It does not replace a factual search tool, evaluation suite, hosted model gateway, or final decision maker.
Technical notes+
pyproject.toml defines a Python 3.10+ Hatchling package with Typer, Rich, a midtry console entry point, strict mypy settings, and pytest configuration. midtry/runner.py detects six executables with shutil.which, builds provider-specific subprocess commands, applies ordered or shuffled Perspective prompts, limits selected CLIs by max_parallel, executes them through asyncio.gather, and returns MidTryResult. midtry/cli.py adds terminal progress, detection, demo, model selection, timeout, concurrency, random assignment, and full or preview output. midtry/__init__.py exposes synchronous and asynchronous library surfaces. tests/test_runner.py and tests/test_cli.py mock subprocesses and cover configuration, filtering, command construction, timeouts, failures, and CLI behavior. docs/ALGORITHM.md specifies scoring, advantage selection, consensus, and final verification, but those mechanisms do not appear in the supplied runtime code, which explicitly asks the user to aggregate responses.
Observed
- License
- MIT License
- Primary language
- Python
- Packaging
- Hatchling package installable with pip; requires Python 3.10 or newer
- Interfaces
- Console CLI and synchronous or asynchronous Python library API
- Supported CLI tools
- Claude, Gemini, Codex, Qwen, OpenCode, and GitHub Copilot
- Test structure
- Pytest suite covers the CLI and runner, including mocked asynchronous subprocess execution
Read from README.md, pyproject.toml, docs/RESEARCH.md, docs/ALGORITHM.md, midtry/cli.py, midtry/runner.py, midtry/__init__.py, midtry/data/__init__.py, tests/__init__.py, tests/conftest.py, tests/test_cli.py, tests/test_runner.py, LICENSE, ROADMAP.md, examples/sheep-test.md.
What it can do
Query multiple LLMs simultaneously with different perspectives
Single prompt or question → Multiple AI responses from different models
Execute parallel CLI commands across multiple language models
CLI command parameters → Concurrent command execution results
Apply different reasoning perspectives to the same problem
Problem statement or query → Multi-perspective analysis results
Compare responses from multiple AI models side-by-side
Query and selected LLM models → Comparative analysis of model outputs
Orchestrate distributed LLM API calls
API endpoints and authentication credentials → Coordinated responses from multiple services
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.