AI Frameworks · xming521

WeClone

WeClone is a Python-based tool that fine-tunes open-source LLMs (like Qwen) using your chat history to create a digital avatar that mimics your communication style. It supports data export from Telegram and deployment to multiple platforms including Telegram, Discord, Slack, and WeChat.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/xming521/WeClone

18.1k

GitHub stars

1.5k

Forks

Python

Primary language

AGPL-3.0

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	xming521/WeClone
Owner	xming521
Primary language	Python
License	AGPL-3.0 — OSI-approved
Stars	18.1k
Forks	1.5k
Open issues	36
Latest release	v0.3.03 (2026-01-04)
Last updated	2026-07-06
Source	https://github.com/xming521/WeClone

What WeClone is

WeClone implements end-to-end LLM fine-tuning via LoRA/QLoRA using LLaMA Factory, with integrated data preprocessing (Presidio-based PII filtering), multimodal support (text + images), and pluggable adapters for chat platforms. Default model is Qwen2.5-VL-7B-Instruct; hardware requirements scale from 4GB VRAM (4-bit QLoRA on 7B) to 1.2TB (full bf16 on 70B).

Quickstart

Get the WeClone source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/xming521/WeClone.git2cd WeClone3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Personal AI Assistant / Digital Twin

Create a conversational bot that replies in your voice across Telegram, Discord, or Slack by fine-tuning on your chat logs. Captures tone, vocabulary, and conversational patterns.

Research/Academic Use of LLM Personalization

Explore how open-source models adapt to individual writing style and domain-specific language through controlled fine-tuning experiments with multimodal input.

Privacy-Sensitive Chatbot Deployment

Run inference locally with full data control; no cloud upload required. Built-in Presidio + blocklist filtering for PII removal before training.

Implementation considerations

CUDA 12.6+ required on Linux; Windows support not rigorously tested (WSL recommended). Mac support assumed but not explicitly confirmed.
Data preprocessing is manual: export from Telegram Desktop, configure Presidio + blocklist in settings.jsonc, run `weclone-cli make-dataset`. Quality heavily depends on chat data volume and diversity; documentation recommends 14B+ models for reliable results.
LoRA/QLoRA fine-tuning times and VRAM usage vary widely by model size and batch settings. Trial-and-error on hyperparameters (`per_device_train_batch_size`, `lora_rank`, `num_train_epochs`) expected.
Deployment adapters for Telegram, Discord, Slack are marked ✅; WeChat (via openclaw-weixin), WhatsApp, Discord full integration are 🚧 (in progress). Verify target platform readiness before committing.
PII filtering via Presidio is best-effort (not 100% guaranteed). Manual blocklist curation is essential for sensitive data; project explicitly warns to verify privacy before use.

When to avoid it — and what to weigh

Strict Commercial/Proprietary Deployment — AGPL-3.0 license requires derivative works to be open-source. Any closed-source commercial product incorporating WeClone must release source code or obtain separate licensing.
Limited GPU Resources — Even with QLoRA (4-bit), 7B model requires ~4GB VRAM; larger or full-precision training demands 16GB+. CPU-only training is not documented as viable.
Non-English or Underrepresented Languages — Default model (Qwen) and filtering (Presidio) are primarily optimized for English and Chinese. Support for other languages or regional dialects is Unknown.
Enterprise Compliance / Data Governance — No formal audit trail, compliance certifications, or SLA documented. Local deployment avoids cloud storage but auditing and governance tools are not mentioned.

License & commercial use

AGPL-3.0 (GNU Affero General Public License v3.0). Copyleft license: any modifications or derivative works distributed online must release source code under AGPL-3.0. Linking against non-AGPL libraries may trigger viral clauses; review carefully.

AGPL-3.0 is NOT a permissive license for proprietary commercial use. Running WeClone as a SaaS service or selling a closed-source derivative requires either: (1) releasing your entire product as AGPL-3.0, or (2) negotiating a separate commercial license with the author. Using WeClone internally for non-SaaS commercial purposes may be acceptable but requires legal review. Do not assume commercial use is allowed without explicit licensing agreement.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Needs review
Deployment complexity	High
DEV.co fit	Good
Assessment confidence	Medium

Security considerations

PII filtering via Presidio (phone, email, credit cards, IPs, locations, IBAN, crypto addresses, age, generic IDs) is best-effort and not 100% guaranteed—manual blocklist review essential. Local fine-tuning and inference avoid third-party data upload. No threat model, pen-test results, or vulnerability disclosure policy documented. LLaMA Factory and dependency security Unknown. Model weights downloaded from Hugging Face; integrity verification not mentioned. Bot adapter credentials (Telegram tokens, Discord webhooks, etc.) stored in settings.jsonc—no encryption at rest documented.

Alternatives to consider

OpenAI Fine-tuning API / GPT-4 Custom Instructions

Proprietary, cloud-managed, minimal setup. No local infra or data preprocessing. Closed-source model; no inspection or privacy control. Requires commercial API key; costs scale with usage. AGPL incompatible.

LLaMA Factory (standalone)

Core training engine used by WeClone; more flexible for model/method selection but no pre-built chat platform integrations. Requires manual adapter development for Telegram/Discord/etc. No PII filtering built-in.

Hugging Face AutoTrain / Replicate

Managed cloud fine-tuning; lower operational overhead. Less data privacy control. Closed-source platforms. Cost per training run. Not tied to specific chat platform integrations.

Software development agency

Build on WeClone with DEV.co software developers

WeClone lets you build a personalized chatbot in hours, not weeks. Grab the GitHub repo, export your Telegram logs, and start fine-tuning. For commercial or enterprise deployment, confirm licensing with the author.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

web-llm-chat LlamaFactory Langchain-Chatchat

Related on DEV.co

Explore the category and the services that help you build with it.

AI Frameworks AI Application Development API Development Cloud Deployment Talk to DEV.co

WeClone FAQ

Can I use WeClone for a commercial chatbot product?

Not without licensing clarification. AGPL-3.0 requires open-sourcing derivative works. A proprietary product would violate the license unless you (1) release source as AGPL, or (2) obtain a separate commercial license from xming521. Consult legal before shipping.

What GPU do I need?

Minimum: 4GB VRAM (QLoRA 4-bit quantization on 7B model). Recommended: 16GB (LoRA on 7B) or 32GB+ (14B models). CPU-only training not documented. Requires NVIDIA GPU with CUDA 12.6+ support; AMD/Mac GPU support Unknown.

How much chat data do I need?

Unknown—no minimum specified. Project recommends 14B+ models for reliable results, implying more data is better. Quality > quantity; diverse conversations capture more nuance than repetitive chats. Experiment with your dataset.

Is my data safe if I run WeClone locally?

Yes, no cloud upload by design. Data stays on-device for preprocessing and training. However, Presidio PII filtering is best-effort (not 100% guaranteed), so manual review + blocklist curation recommended before using sensitive data.

Custom software development services

From first prototype to production, DEV.co delivers software development services around tools like WeClone. Our software development agency staffs experienced software developers and web developers for custom software development, web development, integrations, and ongoing support across ai frameworks and beyond.

Ready to Clone Your Chat Style?

Talk to DEV.co Schedule a Consultation