Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

manyminds

Status: early alpha. The core loop works on real models: paste text, get leads, research them in parallel, read their findings and one report on them all, ask about what was found, and come back to saved checks later. Settings and key storage are still to come. See PLAN.md for what is done and what is next.

What it is

A Python app that helps you check claims you find on the web. It runs on your own machine with your own OpenRouter API key, and its only interface is a local browser app.

Each check is a conversation. You paste a piece of text or a link, with anything you want to say about how to check it. The app replies with leads: specific threads that could be researched to see whether the text is supported by evidence. Press Go on the leads you care about, reword or remove others, or leave them. Each started lead gets its own research agent, which runs alongside the others. When an agent finishes, its findings appear in the conversation: sources with links and quoted passages, and what it could not find out. Once every lead you started has finished, the conversation writes a single report from all of them. Open any thread to see what its agent searched for and read.

The conversation continues underneath. Ask it to summarise what the threads found or to suggest further leads. It answers from the threads' findings first, and can run a quick web search of its own when they do not hold the answer; it says which is which. What it finds that way has no check on its links, as a thread's findings do, so start a lead for anything that matters.

manyminds gathers evidence for you to judge. It does not issue verdicts on what is true. Findings link to their sources, and any link an agent wrote that its searches never returned is listed as such. Quotations are not checked against their sources, so read the source before relying on one.

Where your text goes

Orchestration is local; inference is not. Text you paste is sent to OpenRouter, to the model provider behind each model you choose, and to the Exa search engine, which runs the web searches. The app listens on localhost only.

Checks are saved on your computer only, in your user data folder (on macOS, ~/Library/Application Support/manyminds), in a file only your account can read. Set MANYMINDS_DATA_DIR to keep them somewhere else. Deleting a check in the app removes its text, leads and reports from that file.

Your API key is read from the OPENROUTER_API_KEY environment variable. It is never written to disk, logged or printed. Storing it in the operating system's keychain is planned.

If the text you paste contains links, the app opens them (through OpenRouter and Exa, not from your computer) so that leads can be drawn from what they say. For a YouTube link that is the video's title and transcript.

You can paste a screenshot or other image into the box, or attach one with the paperclip. Attached images are sent to OpenRouter and the model provider, which describes them; they are saved with the check on your computer and deleted with it.

If a link will not open, as posts on X often will not, the app looks for the same material elsewhere, such as a mirror or archive. That step is given the link only, not your words. Images in your text or on the opened pages are shown to the model, which describes what is visible. The reply lists what was opened and what each image was taken to show, so you can check it. A description guides what is researched; it is never counted as evidence. Video is not watched, and a link that needs these extra steps can cost a few cents.

What you paste and what you ask for share one box, so the app cannot tell a request you typed from one hidden in text you pasted. Such text could steer what the conversation searches for or opens. It cannot start research or change anything you have saved.

Research agents search the web and open pages, which can contain text written to mislead a model. What they read is treated as material to examine and their output is checked against a fixed format, but no defence here is complete: a page could in principle steer an agent into opening an address that carries part of your text with it. Read the sources yourself, and do not paste anything you would not want sent to these services.

Run

export OPENROUTER_API_KEY=...     # create one at https://openrouter.ai/keys
uvx manyminds

A lead costs roughly one to two US cents with the default model, most of it the web searches. The cost of each finished thread is shown under its findings.

To look around with placeholder data, no key and no cost:

uvx manyminds --demo

Install

uvx manyminds

or

uv tool install manyminds

Development

Everything runs through uv:

uv sync
uv run pytest
uv run ruff check
uv run ruff format --check
uv run mypy

Development follows strict test-driven development; the rules are in CLAUDE.md.

Licence

MIT. See LICENSE.

Metadata

Release files for manyminds 0.1.0a1

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for manyminds 0.1.0a1
File Size Uploaded Provenance Actions
manyminds-0.1.0a1.tar.gz 65.5 kB Without provenance Details for manyminds-0.1.0a1.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for manyminds 0.1.0a1
File Interpreter ABI Platform Actions
manyminds-0.1.0a1-py3-none-any.whl Python 3 none any Details for manyminds-0.1.0a1-py3-none-any.whl

Total release size: 141.0 kB

Release files / manyminds-0.1.0a1.tar.gz

Download URL manyminds-0.1.0a1.tar.gz
Size 65.5 kB
Tags Source
SHA-256 checksum
How to use checksums
02bb44b808b1691330a3eee8bf4ece1230c808985a7aa156152110e806508bfc
BLAKE2b-256 checksum
How to use checksums
b2abf46be9c0b1b68093c68ad7fec5b80d870eed7f9726edc2e0cbf25cd4bc62
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.13.7 {"installer":{"name":"uv","version":"0.13.7","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Release files / manyminds-0.1.0a1-py3-none-any.whl

Download URL manyminds-0.1.0a1-py3-none-any.whl
Size 75.5 kB
Tags Python 3
SHA-256 checksum
How to use checksums
d22185c1de58bc0b63f2307cd2c7e65feffcc6da85a60ae99c1c18c7251b136e
BLAKE2b-256 checksum
How to use checksums
a730f7525e55072b1d398e32ac1b52574a770596ef9ca38eacbcbd020106d3b6
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.13.7 {"installer":{"name":"uv","version":"0.13.7","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Security for manyminds 0.1.0a1

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page