全模型全渠道附件上传deepseek功能#256

Merged

CJackHwang merged 18 commits intomainfrom

dev

Apr 12, 2026

Commits on Apr 8, 2026

Merge pull request #254 from CJackHwang/main
CJackHwang
authored

Commits on Apr 12, 2026

feat(prompt): align DeepSeek prompt assembly with tokenizer-style turns
CJackHwang
committed
Merge pull request #255 from CJackHwang/codex/refactor-prompt-concatenation-using-tokenizer
CJackHwang
authored
feat: add sanitization for think tags and BOS markers in leaked output and update golang.org/x/net dependency
CJackHwang
committed
feat: implement OpenAI-compatible file upload and reference handling for DeepSeek API
CJackHwang
committed
refactor: fix file handling error suppression, optimize hash calculation, and update API documentation with additional models
CJackHwang
committed
fix: add JSON headers to DeepSeek requests and prevent string content from being parsed as file IDs in OpenAI adapter
CJackHwang
committed
feat: add file attachment support to chat interface and API requests
CJackHwang
committed
refactor: handle upstream thinking-only responses as errors and sanitize dangling think tags in output
CJackHwang
committed
refactor: remove unused purpose parameter from upload and upstream empty output handlers
CJackHwang
committed
feat: update to v3.4.0 and redesign model selection UI with a dropdown and descriptive panel
CJackHwang
committed
feat: implement file readiness polling and add IsImage field to upload results
CJackHwang
committed
feat: implement cross-account validation and improved error handling for file attachments in API tester
CJackHwang
committed
feat: implement error handling for empty upstream responses in chat streams and update UI to display stream-level errors
CJackHwang
committed
refactor: rename apiTester testSuccess key to requestSuccess and update localization files
CJackHwang
committed
refactor: remove unused leakedDanglingThinkOpenPattern regex from output sanitizer
CJackHwang
committed
fix: update URL decoding method and refine file ID extraction logic to exclude text-based inputs
CJackHwang
committed
feat: enforce request body size limits and restrict inline file count to prevent resource exhaustion
CJackHwang
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

全模型全渠道附件上传deepseek功能#256

全模型全渠道附件上传deepseek功能#256
CJackHwang merged 18 commits intomainfrom
dev

Commits on Apr 8, 2026

Merge pull request #254 from CJackHwang/main

Commits on Apr 12, 2026

feat(prompt): align DeepSeek prompt assembly with tokenizer-style turns

Merge pull request #255 from CJackHwang/codex/refactor-prompt-concatenation-using-tokenizer

feat: add sanitization for think tags and BOS markers in leaked output and update golang.org/x/net dependency

feat: implement OpenAI-compatible file upload and reference handling for DeepSeek API

refactor: fix file handling error suppression, optimize hash calculation, and update API documentation with additional models

fix: add JSON headers to DeepSeek requests and prevent string content from being parsed as file IDs in OpenAI adapter

feat: add file attachment support to chat interface and API requests

refactor: handle upstream thinking-only responses as errors and sanitize dangling think tags in output

refactor: remove unused purpose parameter from upload and upstream empty output handlers

feat: update to v3.4.0 and redesign model selection UI with a dropdown and descriptive panel

feat: implement file readiness polling and add IsImage field to upload results

feat: implement cross-account validation and improved error handling for file attachments in API tester

feat: implement error handling for empty upstream responses in chat streams and update UI to display stream-level errors

refactor: rename apiTester testSuccess key to requestSuccess and update localization files

refactor: remove unused leakedDanglingThinkOpenPattern regex from output sanitizer

fix: update URL decoding method and refine file ID extraction logic to exclude text-based inputs

feat: enforce request body size limits and restrict inline file count to prevent resource exhaustion