Skip to content

Releases: SIMON-WORLD/codex-deepseek-vision

Release list

v1.0.4

v1.0.4 Latest
Latest

Choose a tag to compare

View all tags
@SIMON-WORLD SIMON-WORLD released this 04 Aug 15:52
v1.0.4
c3dd26d

v1.0.4

PyPI package is now codex-deepseek-vision (agent-vision was rejected by PyPI as too similar to an existing project).

  • Install: \pip install codex-deepseek-vision\
  • CLI unchanged: \�gent-vision setup\
  • Automatic PyPI publishing via GitHub Actions on version tags
Assets 2
Loading

v1.0.3 - Codex + DeepSeek V4 Flash vision bridge

v1.0.3 - Codex + DeepSeek V4 Flash vision bridge

Choose a tag to compare

View all tags
@SIMON-WORLD SIMON-WORLD released this 04 Aug 12:18
v1.0.3
b9ae777
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

v1.0.3

定位收敛为 Codex + DeepSeek V4 Flash 看图桥（vision bridge）：默认免费 GLM-4V-Flash，不换模型、不用 Ollama/GPU。

新增

  • agent-vision see <图片或URL>：支持 http(s) 图片 URL
  • agent-vision see --latest：从 Codex 会话文件恢复最近粘贴的图片
  • agent-vision see --task describe|ocr|ui|chart，-q 可覆盖
  • 兼容入口 vision_bridge.py 改为薄 re-export，保留 deprecated 提示

健壮性

  • 会话扫描只读最近 20 个文件、尾部倒序分块读取、容错损坏 JSON 行
  • URL 下载校验 scheme / Content-Length / 大小上限
  • images 与 --latest 互斥校验

测试：122/122 通过。

Loading

v1.0.2

v1.0.2

Choose a tag to compare

View all tags
@SIMON-WORLD SIMON-WORLD released this 04 Aug 10:33
v1.0.2
b25d654

v1.0.2 - Codex DeepSeek vision release

Positioned for the primary use case: DeepSeek V4 Flash now speaks the Responses protocol and runs inside Codex/ChatGPT, but the model is text-only. agent-vision adds pasted-image capability by converting images to text locally through a free OpenAI-compatible vision API.

Changes since v1.0.1

  • Fix: setup reloads .env before the immediate vision test, so a clean machine no longer reports VISION_API_KEY not set right after writing the key.
  • Fix: when Codex loads its model list from a local model catalog (e.g. cc-switch), setup declares image input for the active model, backs up the catalog, and restores it on rollback. Pasting images now reaches the local proxy instead of being rejected by the client.
  • Positioning: README now leads with Codex + DeepSeek V4 Flash; OpenCode remains automatic and Claude/Cursor remain guided.
  • Docs: one-line deploy prompt, clickable vision provider links; generated demo screenshots removed.

Verification

  • 107 unit tests passing
  • python -m build + twine check passed for sdist and wheel
  • GitHub Actions tests green on Python 3.9-3.12, ubuntu and windows

Install

git clone https://github.com/SIMON-WORLD/agent-vision.git
cd agent-vision
pip install .
agent-vision setup
Loading

v1.0.1

v1.0.1

Choose a tag to compare

View all tags
@SIMON-WORLD SIMON-WORLD released this 04 Aug 08:31
v1.0.1
16028fc

v1.0.1 - Production hardening release

Give any AI agent vision capability. agent-vision is a universal Agent Vision Adapter: text-only models keep reasoning while a free OpenAI-compatible vision API turns images into text.

Highlights

  • User config directory: all state now lives in ~/.agent-vision/ (Windows %USERPROFILE%\.agent-vision\), overridable with AGENT_VISION_HOME; works identically from source checkout and pip install, no more site-packages writes.
  • Codex safe patch: setup now only rewrites the active provider's base_url to the local proxy; wire_api = "responses", model and API keys stay untouched. The old wire_api = "chat" write that could break Codex is removed.
  • Packaging: MANIFEST.in, SPDX license metadata, project URLs, Python 3.9-3.12 classifiers; sdist includes README/LICENCE/examples.
  • CI: GitHub Actions tests on Python 3.9-3.12, ubuntu and windows.
  • Security: .gitignore now covers providers.json, runtime.json, *.state.json, *.log.
  • Legacy vision_bridge.py kept as a deprecated compatibility entry.
  • 97 unit tests passing.

Install

git clone https://github.com/SIMON-WORLD/agent-vision.git
cd agent-vision
pip install .
agent-vision setup
Loading

v1.0.0

v1.0.0

Choose a tag to compare

View all tags
@SIMON-WORLD SIMON-WORLD released this 03 Aug 18:36
v1.0.0
cecb03b

v1.0.0

Give any AI agent vision capability. agent-vision is a universal Agent Vision Adapter: text-only models keep reasoning while a free OpenAI-compatible vision API turns images into text.

Highlights

  • Runtime Manager: agent-vision start / stop / restart / status
  • One-click setup: agent-vision setup detects the agent, configures the provider, starts the runtime, verifies the vision API and patches Codex or OpenCode with backups
  • Agent adapters: Codex and OpenCode fully automatic; Claude Code and Cursor detected with exact manual steps
  • Pluggable vision providers: Zhipu (free default), DashScope, OpenAI, Gemini, Groq, SiliconFlow, OpenRouter, custom OpenAI-compatible endpoints
  • Bilingual README (English / 中文), agent-driven install guide, real demo output
  • 83 unit tests passing, zero third-party runtime dependencies

Install

git clone https://github.com/SIMON-WORLD/agent-vision.git
cd agent-vision
pip install .
agent-vision setup
Loading