Done Hub: vanhemmat versiot ja versiohistoria

New Features
- User Agreement / Privacy Policy: editable content in admin with independent toggles, consent checkbox at registration and a re-confirmation dialog at sign-in, validated on both client and server
- Automatic log cleanup: configurable toggle and retention period (30 days by default), deleted in batches daily at 03:00
- New request_id and upstream_request_id columns on logs, searchable by either ID
- Prices can be bulk-imported from models.dev
- Long-context tiered pricing: requests exceeding the threshold are billed at long-context ratios for the whole request
- Batch channel deletion
- Language settings: configurable default language, with a prompt when the browser language differs
Models & Providers
- New BedrockMessages channel type (Claude in Bedrock native Messages endpoint), supporting both SigV4 and Bearer auth
- AWS Bedrock now supports the /v1/responses endpoint
- xAI (Grok) image edits
- SSE streaming for image generation / edits
- Lossless passthrough of Gemini grounding metadata, with new maps / retrievedContext sources
- OAuth token refresh now routes through a unified proxy client supporting http / https / socks5
Stability & Billing
- Quota columns migrated from int32 to bigint, fixing large-amount billing overflow into negative values
- New MaxPromptTokens input guard; oversized requests return 400 immediately
- Idle timeout added for streaming responses
- Fixed Origin / Referer passthrough causing Claude channels to be misjudged as 403 and auto-disabled
- Quota reminder threshold adjusted, with a separate toggle
- Misc fixes to channel testing, model mapping, and request IDs
UI
- RPM now consistently uses a 60-second sliding window over the logs table, with TPM added
- New "Fingerprint Passthrough" toggle
- Multi-user token statistics page
- Refined price cards and log details (image tokens, cache writes, long context)
Deployment
- New RELAY_ONLY mode: exposes only relay endpoints and /health; frontend and /api admin routes return 404

Yo long time no see, miss you guys like crazy—this release is absolutely packed with like a million things, frontend redesign, layout tweaks, channel-level header/body passthrough toggle, cost analysis, admin token search, CGO-free builds, Codex now does images, and a bunch of other stuff I can't even remember, so I'm not even gonna try to list 'em all—just go update and play with it, I promise you'll dig it, and huge thanks for all the support and stars, love you all, bye!

Yo update dropped, nobody asked but here we are—retries now let you set cooldown per status code like 503 gets two minutes in timeout jail, 502 gets one, 429 already knows what it did, logs finally tell the real reason you failed instead of just "lol retry_exhausted," graceful shutdown means Ctrl+C doesn't nuke your in-flight requests anymore, Gemini fixed so one dead key doesn't take the whole chain down with it, 504 and 524 actually get a second chance now (408 still dead sorry not sorry), Claude does byte passthrough so all your weird fields survive, image generation auto-downgrades when you accidentally call it via chat so your base64 doesn't turn into a million-token bill that costs more than your rent, connection pools not stuck in 2005 anymore, tokens got billing tags so finance stops crying, admins can transfer tokens across users like Robin Hood but legal, new "trusted employee" role that's better than a normie but worse than an admin, sidebar rewritten so it finally looks like 2026, quota input remembers if you like dollars or tokens, plus some boring Stripe and Redis stuff—anyway update if you want, I'm not your mom, bye.

What's Changed
Right, so—long time no see. Or maybe not. Whatever, no one really gives a toss anyway, do they? Bet you're thinking, 'Right then, must be a massive update, yeah?' Wrong. Just a bunch of tinkering. That said, I did notice while hammering the thing myself—high concurrency, proper meltdown territory—memory and CPU were having a right old barney. All them base64 requests flying about, routing logic’s a proper mess, serialise this, serialise that, memory's just ballooning like it's got nowhere better to be. Luckily, Claude me stepped in. Sorted a load of memory bollocks under the Gemini router, cut down on the CPU spinning its wheels. If you're running a big ol’ pool of Gemini keys, give this version a spin—might save you a pretty penny on server bills. My Alipay's 13543******. Anyway, that's enough of that. Been slow lately—loads of non-critical bugs I just haven’t had the time for. Fancy lending a hand? Pop by the issues page, maybe I’ll chuck some AI credits your way. Right then, catch you next time.
New Contributors
- @yusnake made their first contribution in https://github.com/deanxv/done-hub/pull/52

Heh, been a while indeed — happy early New Year! This update patches a handful of minor issues, including fixes for the Antigravity channel and retry logic in certain Codex call scenarios (thanks @xxxxuanran). Also sorted out a parameter override bug in custom configs (cheers @mengtina), and tried to smooth out some streaming response quirks. Nothing major, just polish really — but if I go too long without a release, you might forget I exist. Then again, we're all bound to be forgotten someday, right? Catch you after the holiday!

Long time no see — and honestly not missing you much! This update is just me neurotically fixing tiny things because I can't ignore details. First, huge thanks to our new contributor @bwwq for integrating the VertexAI Express channel — you're a real one, wish you happiness every day and immortality. Also tweaked a bunch of memory issues for better performance under high concurrency. Tried fixing the Antigravity channel's 429 error… and failed hard. If any expert can help, you too shall be blessed with eternal joy and endless life. Patched some billing quirks in Claude responses and cleaned up a few log prints. That's it — see you next time, and good night.

Here's another update — This round squashes some small issues and adds a rate multiplier to payment gateways so you can customize top‑up ratios per gateway. Might seem niche, but if you're juggling third‑party payments and points platforms like LinuxDoCredit, you’ll appreciate the flexibility. Also fixed Epay callbacks (still puzzling why one‑hub and new‑api handle them so differently…) and cleaned up time‑zone quirks in stats. Take it or leave it — but if it breaks, that’s on you 😉

This is a heavyweight update — just merged code with the one-hub project again (takes hours every time since the two repos are totally independent, not even forked from each other…). Fixed a few of their untested new features while at it, rolled out multi‑user token stats, fully refactored the model pricing module, patched their backup token group bug, and optimized channel sorting cache. Fair warning — this touches a ton of code, so hidden bugs may be lurking. If you spot anything off, report it fast! And remember, your stars are what keep me pushing — don't stop!

Alright — most of the recent small updates were just to adapt the /claude interface for better Claudecode compatibility. Should be all good now. Also polished up some UI here and there. There’s a bigger update coming down the road… just hold tight!

Been thrilled to see how much everyone's loved v1.9.x — seriously, thank you all for the support! Turns out I rushed the last release a bit and didn't test everything locally, so some things broke (my bad). This version patches up routing for /claude and /gemini, polishes some UI bits, and overall just makes things smoother. Shoutout to everyone who's been giving feedback — even if this project doesn't have many stars, you guys keeping me honest is what keeps it alive!

Long time no see — missed y'all! This release finally pushes a bunch of small features into a stable version, while also quietly fixing a few "big" little bugs. Gotta admit, the original Claude cache billing had a serious double‑charge bug (thanks, upstream dev…), so anyone using this project was probably paying extra (lol). Also tweaked some escaping compatibility. Oh, and you'll notice the model pricing page now shows units in "M" by default and only displays available models — you might not like it, but you're gonna accept her, 'cause I do. Oh right, also optimized the retry logic for Gemini channels: if a Gemini channel hits its daily limit, the <model‑channel> pair will auto‑freeze until 4 PM next day, Beijing time.

Missed you guys this past week! This update's packed with optimizations and compatibility tweaks, but let's be real — even if I explain them you'll just zone out and wonder what I'm mumbling about. So I'm staying mysterious this time, go figure out what's new yourself!

This version is absolutely stacked — right after our last big feature, we're rolling out another game-changer: Codex reverse proxy! If you've got an OpenAI Team subscription, you can now proxy it into a full API while keeping the original project's load-balancing goodness. And that’s not even the headline — we also merged upstream (shoutout to the OG devs!) and now your created tokens can restrict access by IP. Oh, and I squashed a few caching bugs too. Enough reading — go test it and let the "NB!" comments fly!

This update's a big one — and yeah, I know the version number didn't bump. Why? Because I barely tested this beast and need you all to be my guinea pigs! So what's new? Fixed some GeminiCli logic, cleaned up the logs, and dropped the bombshell: ClaudeCode reverse proxy support! Now you can proxy its API to use in other clients, share accounts with friends, or do whatever legal mischief you dream up. Not my department! Just help me test this thing before we call it official.

I tweaked the GeminiCli channel config so you don't have to enter a ProjectId anymore. Yep, that's literally the only change in this update — it's so minor you could just skip it. But hey, if you're not even using the GeminiCli channel, maybe this is your sign to give it a try? I mean, who would say no to free Google credits? I'm sure you won't let this chance slip.

This update fixes the fallback mechanism in the token configuration's backup group. And don't ask me why there was a bug in the first place — because I wasn't the one who wrote that feature! Honestly, you should be thanking me for fixing it.
Besides, I bet you didn't even know the token config supported such a powerful yet easily overlooked feature like the backup group. Go try it out now — and learn to pay attention to the details.

When you update to this version, you'll probably feel like nothing has changed from the previous one. But if you really think that — you’re totally wrong.
This release actually optimizes some memory performance issues and high concurrency problems. We even tweaked parts of the UI and interaction design.
But let’s be real — you don’t actually care about any of that. You only care about yourself.

Hey, this update now supports the gemincli channel, which basically means it comes with a built-in 2api. Pretty cool, huh? Damn right it is! As for how to use it... you might have to figure that out on your own. Good luck! Let's gooo!

Don't ask me what's in this update—I can't even remember myself. I pushed the new image after coding but totally forgot to publish the release, so I have no clue what I actually improved. But honestly, it doesn't even matter. Nobody really cares what I write here anyway. Just go ahead and update to the latest version when you see this.

What's Changed
- fix: epay empty value bug by @WormW in https://github.com/deanxv/done-hub/pull/19
- add: add channel api url filter by @WormW in https://github.com/deanxv/done-hub/pull/18
New Contributors
- @WormW made their first contribution in https://github.com/deanxv/done-hub/pull/19
Full Changelog: https://github.com/deanxv/done-hub/compare/v1.7.0...v1.7.1

1. Merge One Hub main branch
2. Support image size settings

1. Add payment callback URL configuration
2. Add chat feature activation configuration

1. Compatible
2. Export Log

Log list style formatting

Fix retry count error

Fixed a bug where channel memory loading failed in certain scenarios.

Attempt to fix channel freeze anomalies under high concurrency

Don't ask me what was changed—I didn't remember.

Refactored the [System Information] feature, an incredibly powerful functionality！

1. Optimized log printing
2. Fixed grouping bugs

Compatible with claude-code

Possibly fixed the coupling issue when simultaneously enabling invite code registration + third-party login.

What's Changed
- feat: 实现模型重定向中“键”自动映射到模型配置 by @AAEE86 in https://github.com/deanxv/done-hub/pull/11
Full Changelog: https://github.com/deanxv/done-hub/compare/v1.4.7...v1.4.8

The reference return for Google Search tool calls under Gemini routing compatibility.

I can't remember what I optimized either.

Fix SQL date formatting

Invitation code

Announcement Rendering Engine Optimization