> For the complete documentation index, see [llms.txt](https://methexis.gitbook.io/methexis-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://methexis.gitbook.io/methexis-docs/readme.md). # Overview ## Overview **Methexis** is a decentralized network for training and governing open‑source AI models.\ Participants contribute **compute** and **datasets**, a stake‑secured validation mesh checks quality, and **on‑chain contracts** coordinate model updates, rewards, and governance using the **$MTHX** token. > **TL;DR**\ > Contribute compute/data → validators verify → model improves → rewards are distributed on‑chain in $MTHX.\ > Ethereum provides coordination and auditability; storage and heavy computation run off‑chain (e.g., IPFS/Arweave + GPU nodes). *** ### Why Methexis? Most existing large language models rely heavily on data scraped from the open web, including forums, social media, and comment sections, often without the explicit consent of users. For example, Reddit itself claims that free access to its content “has already been a foundational part” of how many leading LLMs were trained. That means many users unknowingly contribute to AI model training. Methexis flips that paradigm: here, **contributors are knowingly rewarded** (or slashed, if malicious) for providing data and compute. Our model ensures transparency, consent, and traceability. * **Openness:** AI training should be verifiable and community‑owned, not gated by a few labs. * **Incentives that work:** $MTHX aligns behavior, honest contributors earn; malicious actors are slashed. * **Traceability:** Every accepted dataset and model checkpoint is referenced on‑chain for a transparent audit trail. * **Scalability:** Heavy ML workloads stay off‑chain; the chain coordinates state, rewards, and governance. *** ### Roles at a Glance * **Contributors** — Provide datasets or GPU compute for training. * **Training Validators (compute providers)** — Run training jobs and co‑sign model updates. * **Data Validators (validation workers)** — Stake, sample, and score incoming datasets; challenge bad inputs. * **Stakers** — Bond $MTHX to secure validator roles and share in rewards. * **Governance (DAO)** — Propose and vote on parameters, upgrades, and treasury spend. *** ### $MTHX in One Minute * **Utility:** Staking (security), rewards (contributors/validators), and governance (voting). * **Economics:** Managed emissions with a long‑term curve; vesting and slashing to discourage abuse. * **Treasury:** Funds audits, grants, and shared infra by community vote. > **Note**\ > Final distribution, vesting, and emissions are detailed in **Token → Distribution & Emissions**. Treat early numbers as *proposed* until ratified by governance. *** ### High‑Level Architecture
**Flow:** 1. Contributors submit data / compute. 2. Stake‑weighted **validation committees** screen data (license, anomaly, duplication). 3. **Training validators** run jobs; if results meet consensus/metrics, an update is proposed. 4. **Contracts** record the new checkpoint, distribute rewards, and apply slashing when needed. 5. **DAO** governs parameters (rewards split, stake thresholds, filters) and treasury usage. *** ### Design Principles * **Security first:** Slashing, committee randomness, audits, and transparent execution. * **Verifiable progress:** On‑chain references for accepted datasets and model versions. * **Pragmatic scaling:** Off‑chain compute + on‑chain coordination to keep costs low. *** ### Quick Links * **How Methexis Works** → `/how-it-works` * **Protocol Architecture** → `/architecture` * **Token ($MTHX)** → `/token` * **Running a Validator** → `/validators` * **Contributing Data** → `/data-contribution` * **Governance** → `/governance` * **Audits & Security** → `/security` * **Roadmap** → `/roadmap` > **Disclaimer**\ > Methexis is experimental infrastructure. Nothing in this documentation constitutes financial or legal advice.