# OpenBMB/AgentCPM-GUI

AgentCPM-GUI: An on-device GUI agent for operating Android apps, enhancing reasoning ability with reinforcement fine-tuning for efficient task execution.

Repository: https://github.com/OpenBMB/AgentCPM-GUI
Canonical: https://ross.abutalabs.com/products/agentcpm-gui
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-01-11T08:24:15+00:00

## Health v2 (maintenance only)
Score: 46/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 61, release rhythm 35, longevity 34
- inputs: {"age_days": 477, "days_push": 234, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1407, forks 133 (observed 2026-08-28T04:04:38.393708+00:00)

## What it is
AgentCPM-GUI is an open-source 8B-parameter on-device GUI agent built on MiniCPM-V that takes Android screenshots as input and autonomously executes user tasks. It is fine-tuned for Chinese and English apps and enhanced with reinforcement fine-tuning for improved planning and reasoning.

## Use cases
- automate tasks on android apps from screenshots
- build a gui agent that operates chinese apps like dianping and bilibili
- run an on-device llm agent for phone automation
- ground ui elements like buttons and input boxes from screenshots
- evaluate gui agents on android task benchmarks
- fine-tune a multimodal model for app automation with rft

## When to choose
- you need an open-source gui agent optimized for chinese android apps
- you want on-device inference with a compact 8b model and short action sequences
- you need screenshot-based ui grounding and task execution

## When to avoid
- you need to automate ios or desktop applications
- you lack a gpu for model inference
- you need a no-code automation tool rather than a model and codebase

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, agent-framework, computer-vision, llm-inference, reinforcement-learning
- domain: artificial-intelligence, mobile-development
- platform: python
- tags: gui-agent, on-device-llm, android-automation, screen-understanding, reinforcement-fine-tuning, multimodal, ai-agents, natural-language-processing, android, gpu

## Member repositories
- OpenBMB/AgentCPM-GUI (main) score 46

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:38.393708+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:38:37.527454+00:00, confidence not recorded.
  - readme: https://github.com/OpenBMB/AgentCPM-GUI (fetched 2026-08-28T04:04:38.393708+00:00, sha 06446f86a446)
- Data as of 2026-08-30T08:39:29.467469+00:00.