d8200dfe54
usage: a plan says when its paid time ends — a ChatGPT account the date its sign-in names (only while it is ahead: a token not refreshed since says nothing of now), a GLM Coding plan (ZCode, or a Zhipu or Z.ai key) its renewal or end from the account's subscriptions as ZCode reads them — beside the plan on the Usage page, under the account in the panel, and in the TUI and magpie quota
yetone2026-09-27 23:23:56 +08:00
8dae6aa25b
usage: each call is kept with the session it was part of — the X-Magpie-Session header when the caller sends one, else the session its agent names itself (Claude Code's x-claude-code-session-id, Codex's session_id, OpenCode's) — as "session" in usage.jsonl, and magpie usage lists the sessions that used the most, so several sessions on one model can be told apart
yetone2026-09-27 23:02:36 +08:00
8f34b6d37e
icon: macOS 26 draws magpie's icon light or dark — AppIcon.icon (the bird on a pale tile, or in white on a dark one) compiled into Assets.car beside the .icns, where the .icns alone was darkened whole in dark mode, the black bird lost on a black tile (#117); the app no longer sets a light picture over it in the Dock
yetone2026-09-27 22:55:16 +08:00
36a93974b3
ui: a subscription's window count stays inside its card at every width — each window's one column is its own width rather than its count's, so a long count like ZCode's 74217 / 140000 · 已用 53% no longer runs past the card's right edge, and a count too long for its window breaks at its · with the share on a second line; widening the window puts the counts back beside their names, as the check now measures a count's own parts, not the stacked box that spans the row (#90)
yetone2026-09-27 22:18:37 +08:00
926362db4a
providers: a removed account's Add it back line can be dismissed for good — Don't remind me beside it hides the line across restarts, and the Add sheet offers the account back under Removed from magpie, one click, for when it's wanted again; removing it again brings the line back (#116)
yetone2026-09-27 22:16:15 +08:00
c76af6822f
ui: a subscription's windows side by side put every count under its name once one's doesn't fit beside it, not one up by its name and the next a line below (#90); a provider's models can be picked all at once (those the filter shows) or none, and their list scrolls on its own, not the editor with it (#115); DeepSeek Harness's own models are listed under its name, apart from magpie's DeepSeek provider, which read as DeepSeek twice (#115); Hide emails says when it is on — a struck eye, 邮箱已打码 and the accent, an open eye while emails show
yetone2026-09-27 21:38:25 +08:00
115f0cbd9d
sync: 坚果云 (Nutstore) syncs from the first time — it answers a read in a folder not made yet with a 409 where other WebDAV servers say 404, which magpie took for a failure (#114); a 409 is now no backup yet, and the first sync makes the folder and puts it
yetone2026-09-27 21:01:18 +08:00
b87a3ce360
agents: Qoder is known as Qoder, not Bun — Qoder CLI is a Bun single-file build whose requests carry Bun's User-Agent and nothing of Qoder's, so magpie's provider in Qoder (and Qoder CN) now carries the key magpie-qoder (magpie-qoder-cn), which the gateway knows an agent by before its User-Agent: usage, the routing view and the models it is shown say Qoder. A Qoder wired before with the key magpie is still magpie's and gets the new key on the next sync, as magpie starts
yetone2026-09-27 20:59:38 +08:00
6c89759239
ui: an outcome told while a dialog is open (Test models in a provider's editor) floats over the dialog's scrim at the bottom of the window — in the footer it sat under the scrim, dimmed and blurred out of sight
yetone2026-09-27 20:50:06 +08:00
3d4ae0db47
routing: a Claude account is known once it answers — Claude Code's rate_limit_event (Anthropic's anthropic-ratelimit-unified-* headers: the five hours and the week, used and when they renew) is kept as the account's allowance, standing an hour for a usage endpoint that turns magpie away; and one not known yet goes before those known in smart routing, not after them, where it never answered and so was never known
yetone2026-09-27 20:46:37 +08:00
d6c3baa0b8
cursor: served straight through Cursor's agent API, not cursor-agent over ACP — each request is a Run on AgentService, the stream cursor-agent itself opens, with its sign-in (the keychain's or auth.json's token, renewed by cursor-agent near its end): the conversation goes whole as AI SDK messages the server reads as blobs, the caller's tools as MCP tools the model calls through CallDynamicTool, and the reply streams back text, thinking and calls, the Run closed once the step's calls are all made. None of Cursor's own tools run. Tested live from a sandbox: Claude Code and OpenCode on Sonnet 5 and GPT-5.4 mini read and write files through several tool rounds; parallel calls, a continuation and an image answered
yetone2026-09-27 20:41:23 +08:00
ef13e07e79
ui: an outcome in the footer (Refresh, Test models, a save) is a solid pill in its colour, larger, and an error stays 8 seconds — a result in the window's corner was faint over a desktop showing through, told only by its colour
yetone2026-09-27 20:31:22 +08:00
261f0594de
web: magpie web's usage tells of MAGPIE_WEB_KEY, which keeps one key across runs and browsers signed in for 400 days (#113)
yetone2026-09-27 20:28:39 +08:00
73b650b281
routing: a request a group's rule sent to a member that can't take it fails over to the members the rules after it that match too name, in their order, before the group's others (#112) — two rules for images, the first's model out of quota, now go on to the second's vision model rather than the group's text-only first; and while the first's rests, the second's goes first. The trace names who comes next, and says a turn stays with whoever took it, not the model the rule named
yetone2026-09-27 20:28:38 +08:00
aa08863d12
grok: Codex works on a Grok subscription — the tools Grok's backend turns a request away for (Codex's custom apply_patch, its namespace of agent tools) are left out, and so are web_search's external_web_access and a null content on reasoning handed back, which the backend couldn't decode the encrypted reasoning beside. Tested live: the Codex CLI through magpie counts a file's lines, writes the number and reads it back over several rounds
yetone2026-09-27 20:10:08 +08:00
1d0c9ccf27
settings: a choice's buttons keep their names on one line in a narrow window — the row's description gives way instead, so the Dock's 跟随窗口 and Codex warm-up's 每周和 5 小时 no longer broke a character to a line
yetone2026-09-27 19:50:43 +08:00
2bb10bca6f
providers: Command Code's balance is asked for with its key, as DeepSeek's and OpenRouter's are — how much of the plan's 5-hour and weekly windows is used, then the dollars left on it; and a model list's context_length (Command Code, OpenRouter) is kept as the model's context window, so the agents are told each one's real size
yetone2026-09-27 19:45:18 +08:00
90fd780eef
grok: served straight through the Responses API the grok CLI talks to, not the CLI run behind the gateway — each request goes to cli-chat-proxy.grok.com with the CLI's sign-in and version, as the CLI's own do: the conversation as it is, the caller's tools, images and all, relayed to a Responses client and translated for the rest. The models are the backend's own list, with each one's context window and efforts, and Grok searches the web by itself. A further account signs with its own home. Tested live from a sandbox: Claude Code and OpenCode on grok-4.7 list, read and write files through several tool rounds; an image and a web search answered
yetone2026-09-27 19:41:10 +08:00
4e83dc6be8
web: magpie web knows its version and updates itself as the app does — the badge's restart puts the new version in and runs it in the old one's place with the same arguments and key, so the open page reloads onto it (#111); routing: a gateway served by magpie serve, which has no window, is told apart from another magpie's window, and the routing view says to stop it so this magpie serves the gateway and shows routing live (#110)
yetone2026-09-27 19:39:45 +08:00
a7dd920e8a
tray: the terminal-only build builds again — the menu bar's usage watcher, which needs the tray, is in a file of its own left out of nogui builds; 9cbcb68 had it beside the card's text, which every build has
yetone2026-09-27 19:33:03 +08:00
9cbcb681dd
tray: a subscription's use can sit beside the icon in the menu bar — Settings → Usage in the menu bar picks one of the Usage page's cards, and its windows' use shows there ("42% · 18%", the shortest window first; a key's balance as it reads), with each window and when it starts again in the tooltip, brought up to date every few minutes from the page's own caches, so the vendors are asked no more often. Windows' tray has no room for text and shows the tooltip only. Asked for on Discord
yetone2026-09-27 19:10:27 +08:00
86a0b90580
providers: Command Code's Provider API is a preset — one key reaches its Claude models on /messages and the rest on chat and Responses at api.commandcode.ai/provider, each request sent where the model list's supported_endpoints say it is served (#93); asked for on Discord
yetone2026-09-27 19:04:04 +08:00
41de0d87eb
gateway: count_tokens counts the masked prompt, and a count that cannot be had is told apart — the count Claude Code asks for before every turn went to the vendor as the conversation was written, so Mask secrets was defeated on every one; it now goes masked, the same prompt the turn it counts sends. An endpoint that doesn't implement counting (404/405/501, or a 400 that says so, in English or Chinese) falls back to the local estimate as before; a limited key or a broken connection yields to the next checked key and rests the one that failed, as a generation would, while a client's own cancellation rests nothing; and a real error — a wrong key, a prompt past the model's context, an image it won't take — reaches the agent. The gateway tests are isolated from the machine now (a temp HOME, inert stubs for the sign-in CLIs and the keychain) so a cache that outlives a test can't reach a real login, and the redaction tests' cold starts run in parallel. (#103)
Zhizhen HeandClaude Code2026-09-27 18:58:23 +08:00
ebdf44821f
sessions: Terminal is opened through proc.Command, as every command magpie runs is — os/exec there failed TestNoCommandBypassesProc and kept main's CI red since 78fe45d (spotted by @hezhizhen in #103)
yetone2026-09-27 18:54:09 +08:00
ab76cce91b
providers: a custom provider's icon can come from its website (#12) — From the website, beside Choose a picture, finds the site behind the base URL (its host, then the domains it sits under: api.deepseek.com, then deepseek.com) and keeps the icon its home page names (an SVG first, then the apple-touch-icon, data: URIs too) or its /favicon.ico, like a picture picked by hand; only public https hosts are reached. Icons are fetched as magpie, since deepseek.com turns Go's own User-Agent away
yetone2026-09-27 18:52:49 +08:00
ab7dd01506
codex: /fast works on a routing group too — a group with a ChatGPT account's GPT model in it is listed to Codex with the Fast tier, and the service_tier Codex then sends is kept through the group's routing and passed on to OpenAI's own backends (a ChatGPT account, api.openai.com) only; before, the request was rebuilt without it and Fast did nothing
yetone2026-09-27 17:51:38 +08:00
78fe45dbc3
usage: Sessions — Claude Code's and Codex's own session files read into a list beside the overview: each session's folder, first prompt, models, tokens and estimated cost at list price, filtered by agent and searched; Resume copies the cd and claude --resume / codex resume that picks it up, and the Mac app can open it in Terminal. Only the latest 200 are read, cached by size and time, nothing written to the agents' folders
yetone2026-09-27 17:47:14 +08:00
0ce1888a32
codex: Fast mode through a ChatGPT account — a GPT model Codex's own catalog doesn't list carries the Fast tier Codex gives its GPT models, and the priority service_tier Codex sends for /fast reaches the backend instead of being dropped; other providers' models don't offer it, and other tiers are still left out
yetone2026-09-27 17:42:43 +08:00
b5340c7033
grok, cursor: a run ends with its caller — one whose client left mid-reply (Esc, /clear, Claude Code quit) is stopped at once instead of answering no one, one waiting on tool results that never come is let go after 5 minutes, not 30, and a stop kills what the CLI started too (its own session on Mac and Linux, a job on Windows). A tool that runs longer still gets its answer, from a run told the whole conversation. Runs are stopped with the gateway
yetone2026-09-27 17:42:01 +08:00
c96cfa0ba0
codex: warm-up — a ChatGPT account's next window can be started as soon as the last one resets, by a one-word request to the mini model at low effort, so the 5-hour (or weekly) clock is already running when work begins; Settings → Codex 预热: off, weekly, or weekly and 5-hour. Checked every 5 minutes, only a window seen idle (nothing used, its reset a full span away) is warmed, three tries per reset
yetone2026-09-27 17:36:20 +08:00
8f915a5a23
web: MAGPIE_WEB_KEY takes letters, digits and - . _ ~ only — a cookie can't hold a space, ; or a quote, so the browser was sent a cookie that wasn't the key and turned away with no word why
yetone2026-09-27 17:26:21 +08:00
84ef25805e
web: MAGPIE_WEB_KEY keeps the page's key across runs — for magpie web kept running as a service, where a new key each start signs every browser out; with it the cookie outlives the browser session too (400 days, Chrome's most). Unset, each run makes its own key in a session cookie, as before. A key under 16 characters is refused. (#109)
Arvak2026-09-27 17:26:32 +08:00
e173710982
backup: the library goes with a backup and a WebDAV sync — the instructions, MCP servers and skills with their files, a fifth part, ticked by default (magpie backup/restore --no-library). A backup without keys leaves the servers' secret env and headers empty and a restore fills them from the servers kept here; a skill's file over 2 MB, or past 32 MB in all, stays behind; a path that climbs out of its skill is refused. What it replaces is kept with the library's backups. Tested from a sandbox: exported without keys, changed and removed things, imported — the server back with the local key kept, the skill back as files, run.sh still executable
yetone2026-09-27 17:24:53 +08:00
c025a927d7
library: DeepSeek Harness (dsh) gets the library — its MCP servers as marked inserts of the dsh-mcp-client plugin in every profile's cordis.patch.yml (id magpie-mcp-<name>), a row the user inserted read like any other agent's and one holding a !!js value left alone; AGENTS.md and skills in its folder. SSE servers and names dsh can't take are refused with a reason (#107)
yetone2026-09-27 17:22:03 +08:00
c7b4f9a6db
usage: the page counts on while it is looked at — every few seconds it asks for the counts again and redraws only when they changed, so a request through the gateway shows without a refresh, and the subscriptions' windows are asked again each minute (the vendors' answers are cached behind them). A window in the background or an open menu is left alone. Tested from a sandbox: requests sent with the page open appeared within five seconds, and nothing redrew while nothing changed
yetone2026-09-27 17:16:05 +08:00
3c5f60b1f3
library: Pi's MCP servers go in mcp-adapter.json when pi-mcp-adapter 3 is installed — it reads that file and no longer mcp.json, which it leaves to the MCP Pi is to have itself (nicobailon/pi-mcp-adapter#680); an mcp.json found then is moved over, as the adapter asks, and once mcp-adapter.json is there it is the file. Before 3, mcp.json as before
yetone2026-09-27 17:10:49 +08:00
3597cbbcc3
library: instructions come in sets — Default and any others named, each its own text, one of them the one the agents read; a set is added, renamed, switched to and removed (after a confirm, never the one in use or Default) in Library → Instructions, and a setup snapshot keeps which was in use. Providers: each protocol keeps its own URL, so a saved chat URL no longer shows as the Anthropic one; a URL typed and not saved yet moves to the protocol picked after it, spelled as that one wants it (/v1 dropped for Anthropic, added for OpenAI). Settings: an open WebDAV, export or import form is no longer emptied when the window regains focus. Tested in the browser from a sandbox. Issues #104#105#106yetone2026-09-27 17:05:59 +08:00
59e4f417b2
web: magpie web serves the app's window to a browser, with the gateway beside it — for a computer that can't show the app: no desktop, WSL, a server over SSH, or the CLI-only build, which now carries the page too (what of the app needs Wails is left out of nogui builds). Every request needs this run's key, which the printed link carries and trades for a cookie other sites' pages can't send with a POST; --lan serves it on the network too, with a warning, --addr elsewhere, --no-open leaves the browser shut. In the browser the page opens links and sign-ins itself, names a folder instead of opening it, and leaves out the tray, the Dock and open at login. Tested from a nogui build in a sandbox: no key, a wrong key and a cookieless POST are refused; with the link Agents, Providers and Settings load and a setting saves. Issue #102yetone2026-09-27 16:21:26 +08:00
dad2787f55
settings: the Dock can follow the window — Hide, With window, Show. With window, magpie is in the Dock (and Cmd-Tab) while its window is open and leaves it when the window is closed, so it can be switched to without an icon kept there the rest of the time. Tested on a sandboxed bundle: shown at start, gone on close, back on reopen; Show and Hide behave as before. Issue #101yetone2026-09-27 16:14:52 +08:00
154d071699
routing: a group's effort can be picked per turn by any classifier, not only TypeSafe's Jev — a model that isn't a decision API (Jev Router on OpenRouter, Jev on OpenCode Zen, or any small fast model) is asked in words which of low, medium, high or xhigh the turn needs, in a call of its own beside the intent's, and told what the turn before needed; the routing editor offers every model for it
yetone2026-09-27 16:11:14 +08:00
5bed05f5fb
providers: CherryIN (open.cherryin.ai) is a preset — a New API relay serving chat, Responses and Anthropic's messages at one host, so its models reach every agent with just a key
yetone2026-09-27 15:59:56 +08:00
16c2588432
cli: magpie <agent> <model> says what the config reads after it — the model under the provider it is named by (OpenCode's magpie-relay/… rather than a magpie/… the config has no provider for) — and (unchanged) when it read that already, where it printed the id asked for under magpie/ as if written. Asked on Discord
yetone2026-09-27 15:33:47 +08:00
ce8d0e065c
devin: served straight through Devin's API, not the CLI over ACP — each request goes to the GetChatMessage RPC the devin CLI itself uses, with its sign-in: the conversation as it is, the caller's tools, images and all, and the reply streamed back text, thinking and tool calls as they come. A family's name is sent as its variant at the effort asked for; Adaptive and Fusion, which the API doesn't serve, are no longer listed. The instructions go at the head of the first message, as Kiro's do, since Devin turns away Claude Code's own in its system field. Tested live from a sandbox: Claude Code and OpenCode on Opus 5.5, SWE-2 and GPT-6 Sol list, read and write files through several tool rounds; images and a tool round trip answered on Claude, SWE, GPT and Gemini models
yetone2026-09-27 15:30:00 +08:00
1d4930dd8e
agents: Alma is known as Alma, not ai-sdk — its requests go out with the AI SDK's User-Agent and nothing of Alma's, so magpie's provider in Alma now carries the key magpie-alma and the gateway knows an agent by a magpie-<agent> key before its User-Agent (usage, the routing view, the models it is shown). An Alma wired before gets the key on the next sync, the key alone sent, as a baseURL would have Alma drop its models
yetone2026-09-27 15:28:33 +08:00
bc33064c14
usage: a subscription's windows side by side keep their bars on one line when one window's count wraps under its name — each window takes two rows of the grid (a subgrid), its names and its bar, so ZCode's 5 小时 bar no longer sat a line above Weekly's
yetone2026-09-27 15:28:33 +08:00
6d80295d00
codex: a routing group whose first model is a ChatGPT account's is listed to a signed-in Codex — it was taken for that account's model, which the backend lists already, and left out, so the group was missing from the Codex app's picker and from /model. Issue #100yetone2026-09-27 15:26:50 +08:00
372bf7331b
providers: a model id typed in the add box and not yet Entered is kept by Save, as if Enter had been pressed on it — Save dropped it, so a hand-typed id looked saved ("TRelay saved") but never was
yetone2026-09-27 15:22:48 +08:00
f7e38d0d2e
agents: a magpie on a port of its own (magpie-dev, a sandbox) takes an agent wired to the installed magpie's gateway (127.0.0.1:3425) for that one's, not drift — the two took turns flagging Codex with Apply again, each rewiring it to itself
yetone2026-09-27 14:54:24 +08:00
6cd76567aa
devin: a turn Devin ends because one of its own tools was refused goes on — it sometimes lists the magpie tools, finds the list long and reaches for its own grep or read to look through it, which are off, and Devin then ends the turn (agent_stopped, tool_rejected) as if it were done, so Claude Code showed half an answer and no error. magpie now tells it to carry on with the magpie tools, up to three times, and the prompt says they are already there to call. Reproduced in interactive Claude Code on devin/swe-2 and checked there: the refused read is followed by the rest of the task, files written and tests run
yetone2026-09-27 14:42:53 +08:00
18af9889fb
usage: a subscription window's count on a line of its own starts at the left, under the name — it kept to the right, as it is beside the name, so a short one looked indented by spaces (ZCode's 5 小时 under Weekly's). Issue #90yetone2026-09-27 14:29:01 +08:00
17f289dd4a
gateway: shared on the local network from Settings — Share on local network puts the gateway on every interface at its port, at once and without a restart, so agents on other computers use magpie's models; a request from another machine must carry the API key magpie makes for it (shown with the addresses, copied, renewed in place), and goes on as this computer's own, which need none as before. Off puts it back on loopback. Tested live from a Windows machine on the LAN: no key or a wrong one is turned away (401), the key lists the models and a chat completion goes through to the vendor with the provider's own key. Saving the Settings page no longer drops what other pages keep in settings.json (models' names and levels, who sees which, the network sharing)
yetone2026-09-27 13:48:57 +08:00
532325f7a6
autostart: on Windows, Open at login switched off in Task Manager's Startup apps reads as off, and switching it on in magpie switches it on there again
yetone2026-09-27 12:53:16 +08:00
7c4a5d114c
providers: an OpenCode model is sent on the API models.dev says it's served on — Grok and GPT on /responses, Claude and MiniMax on /messages — so Claude Code asking OpenCode Go for Grok goes straight to /responses; OpenCode's "Model does not support this protocol" is also taken as the model being served elsewhere
yetone2026-09-27 12:51:43 +08:00
0669c450e3
settings: Open at login starts magpie in the tray when you log in, so agents find the gateway from the start — a launch agent on the Mac, the user's Run value on Windows, an autostart entry on Linux, which is where whether it's on is read from too (magpie autostart [on|off] in the terminal)
yetone2026-09-27 12:44:39 +08:00
6cd4344a58
gateway: OpenCode's "Model … is not supported for format anthropic" (and oa-compat) is taken as the model being served on another endpoint, so Claude Code asking OpenCode Go for a Grok model — served on /responses only — goes on to it rather than stopping at the first refusal, and later requests go straight there
yetone2026-09-27 12:34:36 +08:00
08309e8c04
providers: Test models sends each model agents see a tiny request of its own, on the endpoint it's served on and with a key that sees it, and marks its chip with a dot — green with how long it took, red with what the vendor said — so a relay whose other models answer can't hide one that doesn't (magpie provider test <id> <model>… does it in the terminal). Usage has Hide emails too, one setting with Routing's, to screenshot the subscription cards
yetone2026-09-27 12:25:46 +08:00
c3052fe43c
test: magpie model efforts gives a model whose reasoning levels aren't known the ones typed — TestModelNameCmd still expected it refused, as before c1d38ccyetone2026-09-27 11:51:36 +08:00
097f399fc1
usage: Kimi Code's plan windows (its /usages, the 5-hour and weekly allowances) show on the Usage page like the GLM Coding Plan's; the Used / Left toggle sits under an Allowances heading above the subscription cards, away from the period's list-price total it never changed, and a balance says why it is only what is left
yetone2026-09-27 11:45:45 +08:00
6f792a3739
library: RTK's card shows RTK's own mark (the helmet from rtk-ai.app), drawn in the text's colour, in place of a generic terminal glyph
yetone2026-09-27 11:39:59 +08:00
c1d38cc50f
effort: Claude Code can start at max — settings.json's effortLevel keeps up to xhigh (/effort max lasts a session), so magpie keeps max as CLAUDE_CODE_EFFORT_LEVEL in its env, which every session asks for (tested: Claude Code then sends output_config.effort max), and another level takes it away. A model whose reasoning levels aren't known (one models.dev doesn't list, a custom provider's) can be given some of none…max on the Providers page or with magpie model efforts, which the lists agents get and the gateway then take it to have: a request's effort is fitted to them, max sent as max
yetone2026-09-27 11:05:47 +08:00
035dd5b5e3
panel: the tray panel shows every subscription's allowance under the agents — a line per account with its two rolling windows (how much used, or left, as the Usage page is set), its balance, or why it can't be read; the rest stays on the Usage page, and the Usage heading folds it away, remembered
yetone2026-09-27 09:55:40 +08:00
06e173670e
usage: Claude's allowance is read at most every three minutes per account, and a 429 from Anthropic's usage endpoint keeps what was known — the Usage page, the switch list and routing each asked it every minute for every account, and Anthropic turns an account asked that often away, which left Claude's usage blank or erroring. A turned-away account waits as long as Retry-After says (five minutes when it doesn't), a window that has reset since it was read shows as fresh, and one never read says it is rate limited and when magpie asks again
yetone2026-09-27 09:50:54 +08:00
03a01a548d
agent: MiMo Code — Xiaomi's fork of OpenCode, sharing its config shape (#98)
smallnest2026-09-27 08:42:44 +08:00
e0cc477990
providers: an Anthropic base URL given as .../v1 or .../v1/messages, as StepFun's docs show it, is taken as its root — magpie asks the root's /v1/messages, so the version went twice and every message was a 404 with nothing to say why, while the model list, asked at both, still answered; StepFun and StepFun (China) are presets, Step Plan or pay as you go
yetone2026-09-27 08:39:08 +08:00
55afcdd4bb
agents: Qoder CN — qoderclicn, the China site's build of Qoder's CLI, is the same CLI with its own accounts and its settings in $QODERCN_CONFIG_DIR, ~/.qoder-cn by default; magpie wires it as it does Qoder, a magpie custom provider with every magpie model and model.name/model.reasoningEffort, as its own agent
yetone2026-09-27 08:05:12 +08:00
84f001da08
agents: Cline lists every magpie model — its built-in openai-compatible provider knows only gpt-4o, so /model offered that alone and the rest had to be typed in one by one. magpie now puts its models in models.json beside providers.json, as the entry Cline's own migration writes for openai-compatible (provider name, baseUrl, defaultModelId, and the models with their context, output, images and reasoning), which takes the place of the built-in list; the user's entry there is stashed and put back when magpie steps out, and Sync keeps the list current. Tested live with Cline 3.0.65's TUI against a sandbox gateway: /model shows provider magpie with every magpie model, picking one runs the turn on it, and magpie reads the pick back
yetone2026-09-27 07:30:59 +08:00
cae88f031c
agents: Qoder — magpie's models in Qoder's CLI (1.1) as a "magpie" custom provider in ~/.qoder/settings.json ($QODER_CONFIG_DIR): picking magpie/<model> or a routing group writes the provider, with every magpie model, its context, output, images and the efforts it takes, and makes it model.name; the effort is model.reasoningEffort. The comments and entries of the user's settings stay, and stepping out of magpie takes its provider away and puts back the model the user had. Qoder offers custom providers only to a signed-in account whose plan has BYOK. It asks every model for enable_thinking, which OpenAI's API refuses, so that is dropped for openai.com. Tested live: magpie → settings.json → qodercli -p → magpie's gateway → upstream, the reply streamed back
yetone2026-09-27 06:28:14 +08:00
0f7e90b4f2
agents: Cline — its CLI's model and effort can be one of magpie's. A provider of Cline's own is refused when a session starts (cline/cline#14180), so magpie takes Cline's built-in openai-compatible provider and makes it the one in use: the gateway, magpie's token, the model, a User-Agent that names Cline (its requests say only AI SDK), and the effort under reasoning.effort. What the user had there, and the provider in use, are stashed and put back when magpie steps out. Tested live with Cline 3.0.65 against a sandbox gateway: the turn comes back and the usage is Cline's
yetone2026-09-27 06:08:58 +08:00
3651b45e00
library: the skills.sh test counts its fetches atomically — the market asks for the skills' pages at once, so the test server's handler runs on several goroutines and go test -race called the plain counter a race
yetone2026-09-27 04:55:15 +08:00
e136cf9a27
gateway: Gemini PDF and audio files kept on text-only routes (#95)
a90e9ae4e5
library: every skill from GitHub is updated with one click — the Library's skills page has Update all beside the skills in the library (and magpie library skill update with no name does it from the command line): each repository is fetched once however many of its skills there are, the agents are written once, and a skill that couldn't be fetched is said while the rest are updated all the same. Asked for on Discord
yetone2026-09-27 04:21:14 +08:00
54e8bc6e1a
gui: a provider icon always draws — the page's files are served with their own types, not the ones Windows' registry gives, which another program may have changed (an SVG not served as one draws no mask icon), and a mask icon that fails to load becomes the generic one, as a picture of the user's own does. Issue #97yetone2026-09-27 03:44:55 +08:00
cf60231465
gateway: a web search magpie runs for Claude Code is told as Anthropic's own — a server_tool_use block with its query and a web_search_tool_result block with the pages found — so Claude Code's WebSearch counts it and lists its links, where it said "Did 0 searches" with the answer alone. This holds for magpie's search tool on any provider, the Claude subscription's WebSearch and magpie's search tool run inside it (no longer passed to the client as a call), and a Responses provider's own web_search_call, whose sources are asked for. Tested live with Claude Code against a Claude subscription, Grok and Codex: each search comes back with its links
yetone2026-09-27 03:41:53 +08:00
5bad1d117f
gateway: count usage from multiline and final SSE events (#94)
b3f613acb4
gateway: Claude Code with ENABLE_TOOL_SEARCH can search the web — it defers WebSearch and loads it with ToolSearch, whose result is a tool_reference magpie read as nothing, and a Claude subscription's run, told its tools as it began, went on without the tool the client had just loaded, so the model said WebSearch would not load. A tool_reference is now told as the tool being loaded, DeferredToolPlaceholder is left out, and a subscription run whose client now offers a tool it lacks is replaced by one told the conversation with it. Tested with Claude Code on Opus (Claude subscription), Grok and Codex models
yetone2026-09-27 02:53:39 +08:00
d7999aefc7
providers: a balance endpoint named on a provider can be asked with an access token of its own (balance.token), with the provider's headers — a new-api relay's /api/user/self tells the whole account's quota to its access token and New-Api-User, not to a key: magpie provider set my-relay balance=…/api/user/self balance.path='$data.quota / 500000' balance.token=… header.New-Api-User=…
yetone2026-09-27 02:30:55 +08:00
9b1641a493
opencode: a model chosen for OpenCode that a provider of the config's own already sends to magpie's gateway is named under that provider — a layout of one provider per family (magpie-relay, magpie-claude, …) is no longer doubled by magpie's flat provider with the same models, nor the default model moved onto it, by magpie opencode <model>
yetone2026-09-27 02:26:11 +08:00
0f59ca2fc0
routing: a turn a rule sent to one of a group's models stays on it for its tool rounds when another member is on the same account — the conversation was kept to the account that answered, so its next request went to that account's first member: Claude Code's turn Jev sent to codex/gpt-5.5 had its tool results answered by claude/claude-sonnet-5, both served by one provider (as Opus and Sonnet on one Claude sign-in are). And a group whose intents are levels has every turn at one of them however torn Jev is: 继续，再加一个基准测试 after a 复杂任务 came out 0.54 to 0.46, below the confidence counted, and went to the group's first member; a torn turn now stays at the level of the one before, or takes the likelier. Tested with Claude Code through Jev and the real Claude and ChatGPT subscriptions
yetone2026-09-27 02:24:12 +08:00
68408d2d00
gateway: a web search a client offers — Claude Code's WebSearch, Codex's web_search, a Chat client's web_search_options, Gemini's googleSearch — is done for every provider, not only the Claude subscription. A provider that searches by itself is asked to (a Codex account and OpenAI/xAI with Responses' web_search, Anthropic's API with web_search_20250305, OpenRouter with its web plugin); any other model — DeepSeek, Kiro, Grok, Cursor, Devin, a relay — is given a web_search tool of magpie's, which the first provider set up that can search answers with a small model of its (the Claude subscription's Haiku, then Codex, Anthropic, OpenAI, OpenRouter), and the model is asked again with what was found: the client sees one reply and no tool call. A request offering search is no longer relayed as is to a provider that would ignore the tool. Tested live: Claude Code → DeepSeek with WebSearch (2 turns), Responses and Chat clients → DeepSeek, Kiro and Grok, with the Claude subscription, Codex and OpenRouter each doing the search
yetone2026-09-27 02:21:49 +08:00
f2b88ad5b4
jev: a group whose intents are levels has every message in one of them — Jev is asked once per set whether the kinds are levels (a noul, kept), and for such a set each turn is read from a choice among the kinds alone, as with a "none of these" option it put 你好 in none; Jev is also told the kind and the reasoning of the turn before, so 继续 after a 复杂任务 stays complex at its effort. Tested live against System One and OpenRouter: 你好 → nano/low, a deadlock analysis → mini/high (xhigh capped), 继续 → mini, 你是谁 after it → nano/low, Claude Code and Responses requests → the effort picked is what is sent
yetone2026-09-27 01:52:57 +08:00
6fd85894ea
routing: Jev (TypeSafe's System One) as a group's classifier — asked once as a turn begins which of the rules' intents the message is, with how sure it is, and, for a group with effort auto, how hard the turn is to think about; the turn's requests then ask their model for that reasoning, where the agent asked for any. TypeSafe is a decision provider: it serves no conversation and is no group's member
yetone2026-09-27 01:31:01 +08:00
c576153dd7
gateway: a web search asked of a Claude subscription is done — Claude Code's own WebSearch tool asks for Anthropic's web_search, which magpie dropped as a server tool while running claude -p with no tools of its own, so the model answered unsearched. The run now has Claude Code's WebSearch when the client offers web_search; its call and result stay inside the turn, the client hearing one reply whose usage counts both requests
yetone2026-09-27 01:39:22 +08:00
a5a8d29ba4
providers: a model list that says which endpoints each model is served on (supported_endpoints, as Command Code's does) is heeded for every provider, not Copilot's alone — a live-fetched list left it out, so Claude Code's request for Command Code's deepseek went to its /messages, which serves only its Claude models, and was turned away; it now goes to /chat/completions, translated, and a Claude model asked for over Chat goes to /messages. /messages and /v1/messages both name Anthropic's. Issue #93yetone2026-09-27 01:29:33 +08:00
4f22d103d3
providers: Kiro is served through Kiro's own API, as pi-provider-kiro does, not kiro-cli over ACP — magpie reads the sign-in kiro-cli or the Kiro IDE keeps (Google, GitHub, AWS Builder ID / IAM Identity Center, or an external IdP), or takes a Kiro API key saved on the provider, and calls generateAssistantResponse itself: the caller's system prompt, whole history, tools, tool calls and results and images go to the model as they are, with no agent's own prompt and tools between, and its event stream comes back as text, thinking (asked for with Kiro's thinking mode), tool calls streamed as they are written, and usage. kiro-cli no longer has to be installed. A token near its end is refreshed — by kiro-cli when it is its sign-in, else directly, the new one written back where it came from — and one Kiro turns down is refreshed once and the call tried again. History is made the shape Kiro takes: it opens and ends with the user, every call is answered in the next message (an interrupted one said to be so), orphan results go, tools the conversation used but no longer offers are described, ids Kiro rejects are hashed, only the latest images are sent again. Kiro's usage limit is a 429, a prompt too long the error the rest of magpie takes for it. The models are Kiro's list with their context and image support; the account is its email and plan; the Usage page shows its credits and free trial. Issue #90yetone2026-09-27 01:25:37 +08:00
4f0b889ec1
usage: TestSummarizeTellsPlacesApart holds at any hour — its calls of hours before were yesterday's when it ran after midnight, and Today left them out; it now runs at noon
yetone2026-09-27 01:25:17 +08:00
cf3da44807
usage: a narrow subscription card puts a window's count under its name — at the default window size ZCode's 5 小时 was squeezed a character a line and its count ran past the card's edge; the name keeps one line, the count takes the next when both don't fit, and one too long even for that ends in … with the whole of it on hover. Issue #90yetone2026-09-27 00:58:09 +08:00
f12a75899d
providers: a subscription whose models come from its own CLI (Kiro, Devin, Cursor …) has no host, so magpie provider models said its models were fetched "from" nothing; it names the provider. Kiro signed in with Google or GitHub says so, not SocialGoogle. Checked against a real Kiro account: plain and streamed text, a tool call and its result, OpenAI chat with another model, and Claude Code reading a file through Kiro
yetone2026-09-27 00:51:17 +08:00
050c33b07d
providers: Kiro — a Kiro subscription served through the genuine kiro-cli over ACP, as Devin's is: magpie runs kiro-cli-chat acp as an agent of its own written into an empty workspace, whose only tools are the caller's (the magpie MCP helper, trusted; includeMcpJson off), picks the model with session/set_model, streams text and thinking back, and refuses permission for anything but the caller's tools. The account is whoever kiro-cli whoami says is signed in; a Kiro API key saved on the provider goes to the CLI as KIRO_API_KEY, the user's own environment's never. Missing CLI, signed out, MCP turned off by Kiro and ACP errors' data are each said plainly
yetone2026-09-27 00:38:10 +08:00
69b295e28a
providers: Kimi Code — a Kimi Code membership's own endpoints (k3, kimi-for-coding …), which Kimi lets members use from other tools with a key from its console, as presets for its global site and its Chinese one, their models from models.dev's kimi-code-plan catalogs
yetone2026-09-26 23:38:46 +08:00
6694108bf2
agent: an agent whose own config keeps the reasoning effort its sessions start with gets an effort field, as Codex, Grok and Pi have — Claude Code's effortLevel (low to xhigh; max lasts a session only), Hermes's agent.reasoning_effort, omp's defaultThinkingLevel, Goose's GOOSE_THINKING_EFFORT, Copilot's effortLevel (the model's levels when its list says), Crush's large model reasoning_effort, Command Code's reasoningEffort for the current model (it keeps one per model) and dsh's llm-deepseek reasoningEffort (off, high or max, kept when the model changes or the catalog syncs); OpenCode, Gemini CLI, Cursor, ZCode, Devin and Alma keep none of their own in a file, so they get none. Asked on Discord
yetone2026-09-26 23:29:07 +08:00
45a5041a20
models: a name is given only to a model its provider has — magpie model name deepseek/nope X was kept and never shown; and the level none reads 不思考 in Chinese, as the other levels are named. Issue #92yetone2026-09-26 23:29:40 +08:00
47e12df418
models: a provider's model can go by a name of the user's, and offer only some of its reasoning levels — Claude Opus 5.5 · Claude Code can be Opus 5.5, and a model with six levels can offer just low, medium and high. Both are kept in settings.json by provider/model (modelNames, modelEfforts), apart from the vendor's list, so a Refresh, a restart or an upgrade leaves them, and the same model from another provider keeps its own; a provider renamed takes them along. They are applied where the catalog is made, so the gateway's /v1/models, the lists magpie writes into the agents (Codex's model catalog, ZCode, Pi, OpenCode, OMP …) and magpie's own pages all show them, and a change rewrites the agents' lists as any other change of the catalog does. Only what is shown changes: the id, where requests go, and the levels the vendor is asked at stay as they were; a found routing group keeps the model's own name. The provider's editor has Names & levels under its models — a name typed in place, a tick per level, Restore default — and the terminal has magpie model name <provider/model> <name>|--reset and magpie model efforts <provider/model> low,medium,high|--reset. Issue #92yetone2026-09-26 23:23:02 +08:00
8630e4c62a
tui: the routing page (3, was groups) does what the app's Routing view does — an open group lists its rules under its models, one list: n adds a rule typed as magpie group rule add takes it (use=<model> tokens=200k images effort=… agents=… intent="…" classifier=<model> at=<n>), ↵ types one again in place, J/K moves it, d twice removes it; c picks the classifier, s how long a conversation stays, x the context agents are told, R renames the group (the page follows it); a found group that was removed is listed last and u brings it back; smart routing is shown as smart, not in order, and o goes round all four. The library page adds too: a adds an MCP server (a URL, or a command) or skills from a GitHub repository or a folder, e changes a server (its env, headers and agents kept), u fetches a skill from GitHub again, s syncs the library into the agents. The rule parsing moves to provider (ParseRule, RuleWords, Rule.Line) and a typed server to library (ServerOf), for the CLI and the TUI alike. Asked on Discord
yetone2026-09-26 23:08:33 +08:00
c0b296b300
agent(zcode): a magpie model tells ZCode how long a reply may be and how hard it can think — ZCode's model dialog showed 32000 max output tokens and thinking only on or off for every magpie model. config.json's models now carry limit.output and reasoning {enabled, variants, defaultVariant} (medium where the model has it, else its middle level; none is ZCode's disabled), and provider_config.json's rules, which ZCode 3.14 reads, carry optionSpecs {maxOutputTokens.max, reasoningLevel.values}, as ZCode's own rules do. An output over 128000 is capped: some vendors report their window there (grok's 500000). Asked on Discord
yetone2026-09-26 23:04:14 +08:00
c08231ee25
usage: a window reads as how much is used or, as the vendors' own apps show it, how much is left — Used / Left on the Usage page, or a click on the figure, flips every meter and its bar, kept for next time; the label says which (64% used, 36% left). A window with the vendor's own count shows the percentage too (345 / 28000 · 1% used, issue #90). The terminal's usage page shows the accounts too — subscriptions' and plans' windows as meters, keys' balances — asked of the vendors when it opens and on r, u to flip used / left
yetone2026-09-26 23:03:23 +08:00
1d064efe4d
provider: TestContextSetByUser gives the config dir a place, so the Devin CLI a saved group asks writes its config there, not into internal/provider/devin
yetone2026-09-26 23:02:05 +08:00
97d4b3154e
gateway: a model's id without a provider in it that names a routing group is served by the group — gpt-6-sol goes where group/gpt-6-sol goes, its cheapest member first and the next when one fails, where it went to the first provider that serves it with no failover; and a found group is named however a vendor spells its model, so grok-4-7 (the group of grok-4.7) is served rather than unknown. With no group it is one provider's as before, a provider/model id is always that provider's, and the Routing view shows the group it went to. Asked on Discord
yetone2026-09-26 23:00:31 +08:00
796dac8de9
agent(pi): a magpie model tells Pi how long a reply may be — maxTokens from the model's output limit, where Pi otherwise capped every reply at 16384 tokens (#91)
yetone