Third-party provider reasoning-effort AND input-modality settings for DeepSeek Harness: thinking levels and image-input support declared per model, auto-adapted from a model knowledge base + wire-protocol inference, edited right inside the official Models
The plugin will be installed here. Keep web if you are unsure.
npx -y @deepseek-ai/dsh plugin --profile web add github:HaoyueQin/dsh-better-reasoning-effort#833c6e0c96716412c1d74a22af4c9286cc7fcd3c
Compatibility and provenance
Better Reasoning Effort is published as dsh-better-reasoning-effort and currently resolves to version 0.4.0. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.
Reasoning-effort and input-modality editing for third-party models in DeepSeek Harness — thinking levels and image-input support declared per model, auto-adapted from a model knowledge base + wire-protocol inference, edited right inside the official Models page card. Plus a quick reasoning-effort slider inside the official model menu (white round thumb, integrated from HanaAyane's dsh-reasoning-effort — see Acknowledgements) — the composer's official bottom-right model · effort display is left untouched.
The llm-pi-ai adapter of DeepSeek Harness natively supports per-model reasoningEfforts declarations (which thinking levels a model accepts, and the exact string to send on the wire for each). But the official Models page editor deliberately keeps this field out of reach — the official notes say it is a per-model capability and a provider-level knob would break some models. As a result:
Third-party models get no thinking-level picker in the composer (getSupportedThinkingLevels short-circuits to ["off"]);
Only the official DeepSeek API (the built-in catalog) can set reasoning effort;
Setting levels for a third-party model meant hand-writing the reasoningEfforts / compat blocks in settings.yaml.
Hand-declared third-party models are treated as text-only (input defaults to ["text"]): image attachments are refused before they are sent, the read-image tool refuses, and every gateway path in between gates on the same flag. The core already accepts a per-model input: ["text", "image"] declaration — the official page just does not expose it either.
This plugin brings both configuration surfaces back into the UI: edit right inside the official model editor card, plus one-click auto-adapt.
Features
In-page injection: an editor block appears in the official Models page under each model row's disclosure, next to context window / max tokens — not a separate settings page, but part of the official editing flow (same settings.mutate contract, same save style). The block spans the full row; its level rows split into the same two columns as the official capacity pair. It now carries two sections — Reasoning effort and Input modalities — whose changes commit with the official card's own Save, with the editor's Reset underneath them.
Pending until you save the card (zero writes while editing): the editor also appears on rows that are not saved yet, through one pipeline for two shapes — a provider's create card (auto-adapt reads the typed protocol/endpoint straight off the card), and a new model row added under an already-saved provider (auto-adapt then works from the stored route facts plus the display name typed on the row). A change is pending the moment you make it, and it lands together with your Save in that card — the official card freezes its own settings revision while it is open, so this plugin commits on the same Save instead of racing it (the "configured it, saved it, and it is gone" report). Cancel (or a reload) discards it together with the card's own fields. A declaration already in the document is never overwritten. Note staging cannot express a deliberate "unset": applying an all-clear draft simply withdraws the staging and lets host auto-fill fill its suggestion back in — to declare nothing durably, save the row first, then clear every level and Save.
Input-modality declaration: one checkbox ("Image input") turns a hand-declared model vision-capable end to end — composer attachments, the read-image tool, and proxy gating all key off the same flag. Unchecking narrows the declaration to text-only; clearing it writes a durable inputUnset marker that host auto-fill respects, exactly like its reasoning-efforts sibling. On kernels whose official Models page already ships its own Input-types editor (0.1.6-alpha.2+), this section self-disables so there is exactly one modality surface — the capability is sniffed from the row DOM, never from a version number; older kernels keep it.
Endpoint-compatibility controls: the editor carries a third section, rendered only on the protocol whose compat gate takes it. On openai-completions routes: the thinking budget field (which parameter carries the thinking-token budget — some vLLM / self-hosted endpoints read thinking_token_budget, others thinking_budget or thinking_budget_tokens; unset sends none) and the (scheduler priority for endpoints started with ). On routes: — some Responses gateways reject the parameter, so "Omit it" keeps an output cap out of the request entirely. Controls use the official field shape (caption above, the official enum width, a hint line below) and each one says what it does. These switches are per-endpoint passthroughs, not model facts: the knowledge base deliberately never predicts them (no model is to reject ), so Auto-adapt will not fill them — set them once for a gateway that needs them. Picking "Unset" and saving the card really removes the key again rather than keeping the last choice; the editor only ever deletes the fields it showed, so a compat field you wrote into by hand stays put.
Install
Requires DeepSeek Harness 0.1.5-alpha.1 or later (the current 0.1.x kernel release line; peer ranges @deepseek-ai/dsh-api-remotes@^0.1.5-alpha.1 || ^0.1.6-alpha.1, @deepseek-ai/dsh-settings@^0.1.5-alpha.1 || ^0.1.6-alpha.1, @deepseek-ai/schemastery@^3.18.0). The peer range is a per-line union rather than >=0.1.5-alpha.1 because semver's prerelease exemption only covers the same major.minor.patch tuple — a >=0.1.5-alpha.1 range matches no 0.1.6 prerelease at all).
On an older DeepSeek Harness? This release line targets 0.1.5-alpha and later — the 0.1.2-rc / 0.1.3-alpha lines and earlier are no longer supported. Please upgrade Harness, or install an older plugin release that matches your kernel (for example dsh-better-reasoning-effort@0.3.7 for the 0.1.2-rc / 0.1.3-alpha lines).
Compiled and gated against 0.1.6-alpha.2 (typecheck, the test suite and the full build all run against that release's official packages); the most recent live-kernel re-check was 0.1.5-rc.1.
Seam-by-seam source re-check across 0.1.5-rc.2 → 0.1.6-alpha.1: the settings service (get / describe / update plus settings/updated) and the generated Typert ctx.remote.settings contract, the two settings.models slot seats, the slots / locale runtime, the connection service and its connection/reset event, the six Models-page anchor aria-labels and the structural class names, the composer model popover (aria-controls → role="menu" → menuitem / menuitemradio), the dsh.client load rules and the /plugins/<id>/client.js route, the llm service wrappers (prepareCall / stream) and webServer.register, and pi-ai's config.ts / catalog.ts (compat keys, reasoningEfforts, input) — every carrying implementation is unchanged. Two neighbouring edits stay off this plugin's injection path: ui-settings-models gained a placeholder and hint copy for deepseek-family endpoints, and ui-input-trigger changed how slash-command menu rows render (this plugin injects into the ModelSelect popover). The 0.1.5-line compatibility notes still hold (verified source-level: the settings Remote wire, the Models-page anchors, the model-directory types, slots / locale are unchanged since 0.1.2-rc.1; only the llm-pi-ai compat schema grew — pi-ai 0.85.1 adds thinkingTokenBudgetField / vllmPriority / supportsMaxOutputTokens — and the composer model menu is now portaled to document.body, which the slider follows through the trigger's aria-controls link with the inline shape kept as fallback). New-schema keys are suggested where the protocol takes them and stripped automatically on a write refusal from an older kernel, with no version sniffing. Seam details: the settings Remote is the generated Typert ctx.remote.settings stub (argument-less describe, positional mutate(ns, ops, expectedRevision), {ok, value | error} envelopes, settings/conflict / settings/rejected refusal codes), the Models-page anchors (Capacities/容量, Model ID, Display name, Provider ID, Base URL, API protocol; the settings.models.footer slot) are unchanged, and the raw-listing probe mirrors the kernel's own model discovery — the same protocol set (now including via its native route with + ), the same dual / listing shapes, and the same 4 MB ceiling. The client bundle requests no official module at runtime, so it loads unchanged.
A further source re-check across 0.1.6-alpha.1 → 0.1.6-alpha.2 found exactly two carrying changes, both now adapted without version sniffing: the sessions list snapshot dropped its current selection (navigation moved to the view owner), so the composer slider resolves the current session from ctx.uiSession's main-view binding, falls back to the first main-view-retained catalog row, and still reads current on older kernels; and ModelDirectory.select now RESOLVES a {ok:false} refusal result instead of throwing, which the effort-memory and slider commit paths normalize alongside the thrown shape. The Models-page disclosure label was renamed from Capacities/容量 to Model options/模型选项; the injector anchors by the modelAdvanced dictionary key and follows it automatically, with Capacities kept as the no-dictionary fallback. 0.1.6-alpha.2 also added an official per-row Input-types editor (ModelInputTypes), which this plugin treats as a capability: the injector sniffs that control in each row's disclosure DOM and hides its own modality section on exactly those rows, so one modality surface remains (older kernels keep the plugin's). Everything else — the settings Remote, the slot seats, the composer menu DOM, the pi-ai schema, the llm wrappers and webServer.register — is unchanged.
One DOM-bypass path for the per-model editor (no version sniffing): the injector keys off the official disclosure anchors by the modelAdvanced dictionary value (Capacities/容量 on the 0.1.6-alpha.1 line, Model options/模型选项 from 0.1.6-alpha.2), so the editor mounts under every model row that expands — inside the edit → custom settings flow — including unsaved rows on a provider's create card (staged, flushed the moment the row is saved). The slider toggle rides the official settings.models.footer slot, declared through the plugin's own remote.settings inject — the same service contract the official Models page consumes. The Models page's other sanctioned seat, the keyed settings.models.provider-card (per provider card), is the migration path for card-level UI — but no slot reaches a single model row, which is why the per-model editor keeps the DOM bypass.
From npm
# under the dsh web profile
dsh plugin --profile web add dsh-better-reasoning-effort
From GitHub
# under the dsh web profile
dsh plugin --profile web add github:HaoyueQin/dsh-better-reasoning-effort
The github: source only pulls source; lib/ is built by the package's prepare hook. pnpm does not run build scripts of git dependencies by default — the installer prints the allowBuilds key it needs; follow that and add again.
Local development
npm install && npm run build
dsh plugin --profile web add link:D:/Project/dsh-better-reasoning-effort
Restart dsh web, hard-refresh the browser. Each model row's disclosure on the official Models page now carries a "Reasoning effort" block.
Usage
Configure a third-party provider (API key etc.) on the official Models page.
Expand a model row: the editor block sits under the official capacity fields.
Check levels (off / minimal / low / medium / high / xhigh / max) and fill the wire values (e.g. give high the spelling ultra, and the gateway receives ultra when you pick High in the composer);
Toggle Image input under Input modalities to declare what the model accepts (unchecked with no declaration = inherit the provider default, usually text-only);
Click Auto-adapt to fill recommended levels and modalities from the knowledge base / protocol / endpoint listing — reference capacities show up as read-only hints you can copy into the official fields yourself;
Any change you make is pending: the block says so, and those changes land when you press the card's own Save. Press Cancel (or reload the page) and they are discarded along with the card's own fields.
On a compatible protocol, the Endpoint compatibility section appears at the bottom — set the thinking budget field / vLLM priority on openai-completions, and max_output_tokens handling on openai-responses.
All levels off + Save = unset the declaration; only off checked + Save = disable reasoning (false); Clear declaration on the modality row + Save = back to inheriting the provider default.
Declared models are immediately selectable for reasoning effort in the composer's model picker, and image-declared models accept attachments end to end.
Configuration
The host half accepts optional configuration on its profile row (the values below are the defaults):
- insert:
- id: dsh-better-reasoning-effort
name: dsh-better-reasoning-effort
config:
# Auto-fill undeclared models on boot (the browser half fills session
# additions once you leave the editing card).
autofill: true
# Whether the auto-fill above also fills input-modality declarations.
modalityAutofill: true
# Upstream /models probe fetch timeout, in milliseconds.
probeTimeoutMs: 15000
# Boot-fill retry backoff schedule; [] means "try exactly once".
bootRetryDelaysMs: [1000, 2000, 4000, 8000, 16000, 30000]
# Map effort-less calls on forced-thinking ladders to the vendor default.
defaultGuard: true
Set autofill: false to disable the silent auto-fill entirely (both the boot pass and the browser half's running complement) — the browser-side Auto-adapt button keeps working.
How it works
Browser (lib/client.js) Host (lib/index.js)
├─ DOM injector └─ Auto-fill
│ MutationObserver on the models page settings/updated → adds a
│ → mounts EffortEditor in each recommended reasoningEfforts
│ model row's disclosure for undeclared models
├─ Composer injection
│ MutationObserver on the document
│ → ComposerSlider (root pane)
│ → model search box (model-list pane)
├─ EffortEditor (React component) (knowledge base + inference)
│ level checkboxes / wire values /
│ input-modality toggle /
│ auto-adapt (zoned suggestions) / committed with the card's Save
│ └─ writes settings.mutate (llm-pi-ai)
Knowledge base + protocol inference: suggestEfforts() in src/knowledge.ts, a pure function shared by host and browser — fusing endpoint signals, curated entries (levels, modalities, reference capacities), a name heuristic, and protocol inference.
DOM injection: reconcile() in src/client/injection/models-page-editor.ts locates model rows by the official button aria-label (the modelAdvanced dictionary value: Capacities/容量 on 0.1.6-alpha.1, Model options/模型选项 from 0.1.6-alpha.2) and mounts the editor into the capacity disclosure. The browser half's assembly layer is src/client/index.ts; each injection seam lives in src/client/injection/.
Writing: createEditorApi() in src/client/ops.ts rewrites providers.<route>.models[i].reasoningEfforts — and, when an intent travels, .input — via settings.mutate, preserving every other row field; on a revision conflict it re-reads and retries once (the same recovery the official settings form uses).
Shared constants: src/constants.ts carries the plugin id, settings namespace, and DOM marker used by both halves.
Development
npm run typecheck # tsc strict check on src
npm test # vitest: knowledge / inference / autofill / DOM injection / writing
npm run build # lib/*.js + lib/client.js (module-loader bundle)
Contract version: @deepseek-ai/dsh-api-remotes@0.1.6-alpha.2 (client contract types; peer range ^0.1.5-alpha.1 || ^0.1.6-alpha.1). The dev dependencies are unified on the published 0.1.6-alpha.2 packages — typecheck (0 errors), the test suite (20 files / 390 tests passing) and the full build all run against them — while the runtime baseline is 0.1.5-rc.1: 0.1.6-alpha.1 and 0.1.6-alpha.2 have had the source re-check and the gates above, not a live-kernel run.
Runtime re-check against the 0.1.5-rc.1 kernel (2026-09): the settings Remote's describe / mutate(ns, ops, revision) contract, the Models-page anchors, and the slider's menu discovery are unchanged; rc.1's two llm-pi-ai tightenings are covered here — a model-level compat key must belong to the protocol that model resolves to (the write path strips per protocol and retries, with the refusal prose pinned verbatim in a test), and stored profiles that no longer validate surface as a row-level error on the provider card instead of failing the whole page. The suite pins composer-menu discovery across the portaled (0.1.5) and inline menu shapes; the 0.1.2-rc.1 downgrade-retry path is kept as a safety net.
Known limitations
Injection depends on the official Models page's current DOM (aria-label/class). If an official upgrade changes the structure, injection pauses until adapted; the official page is unaffected meanwhile.
The official model menu's Arrow-key roving focus walks its own (hidden) root cells, which is a no-op on display:none nodes — keyboard users reach the replica via Tab, and the replica row's Enter opens the official model list.
The auto-adapt probe route answers loopback and IP-literal Hosts only — the core /api fence's Host-allowlist discipline without its trustedHosts escape hatch (a rebound page always names the attacker's domain in Host, so named hosts are refused outright). LAN deployments serving the GUI under a domain name get a 403 from this one route (IP-literal LAN hosts keep working); every other feature is unaffected.
The auto-adapt probe never follows a redirect (redirect: 'error'): it carries your stored credential, and a cross-origin hop would not strip the custom authentication headers it composes — so only the address named in the profile may receive it. The cost is bounded: a gateway that lists its models only behind a 30x yields no endpoint evidence, and Auto-adapt falls back to the knowledge base / protocol inference, the same path every unanswerable endpoint takes.
reasoningEfforts declarations are suggestions: which levels/spellings an endpoint actually accepts is up to its docs — tweak each in the UI.
The knowledge base is not exhaustive — spellings drift as vendors ship models, and families without an effort ladder carry no entry at all; unlisted models fall back to protocol inference + generic levels and can be adjusted by hand.
Endpoint-compatibility switches are never auto-filled, by design: supportsMaxOutputTokens and vllmPriority describe a gateway's behavior rather than a model's capability, so no model entry carries them. The safe default (unset) sends the field the protocol normally sends; flip the switch only for a gateway that has actually refused it.
The modality vocabulary follows pi-ai's core (text / image today). Wider support some gateways serve (PDF, audio, video) is recorded per family until the core vocabulary grows — declaring them is impossible today by design, not oversight.
For a model declared image-capable, how an over-budget image request fails depends on the kernel: 0.1.6 fails with IMAGE_OFFLOAD_REQUIRED and lets the compaction side choose which images to offload, where 0.1.5 silently replaced the oldest images with placeholder text. The declaration and the editor are unchanged — this only affects the shape of an over-budget request's failure.
Name-heuristic modality advice (vision-flavored ids like / / ) is deliberately low-confidence and labeled as such — verify before relying on it.
Acknowledgements
The composer reasoning-effort slider is adapted from dsh-reasoning-effort by HanaAyane (MIT license) — thank you for the original work and the codex-style effort control idea.
What this plugin took from it:
the session model-selection contract it rides (per-session model directory → adapter-advertised effort ladder → selectModel submit, with optimistic snap and rollback on refusal);
the slider interaction shape (drag / keyboard, level label next to the thumb).
What was deliberately changed in this integration:
White round thumb only. The chibi-runner "big fish" knob is not carried over (it swaps the thumb for the fish sprite); everything else is upstream verbatim — the gradient pill track, the left-clipped radiation canvas effect and the flare glow, the drag/keyboard contract, the optimistic commit with rollback.
The official model seat is never replaced. The upstream plugin shadows the whole seat (its own trigger + menu); here the official bottom-right model · effort display stays untouched, and the slider is injected into the top of the official menu when it opens.
Different placement / fewer settings. The upstream "推理强度滑块 / 大肥鱼滑块" items lived in the general settings page; here only the Reasoning effort slider toggle remains, in a boxed container on the Models page below the add-provider actions. The "大肥鱼滑块" item is dropped together with the feature.
Maintained on the 0.1.5-alpha line and later (compiled and gated against 0.1.6-alpha.2). This is a reduced re-implementation over the harness wire contract (not a fork of the upstream bundle): it runs on 0.1.5-alpha.1 and later kernels (see the compatibility note above) without the upstream's 0.1.0-rc.6 pins, and the whole mount/unmount lifetime is managed by this plugin's DOM injector. If the upstream project resumes publishing, keep both in mind: running both plugins doubles up — the upstream shadows the official seat again, so the official trigger would disappear once more.
If you used the upstream plugin before, remove it to avoid two effort controls on the same seat:
dsh plugin --profile web remove dsh-reasoning-effort
Zoned suggestion display: Auto-adapt reports what it applied (source · confidence) on its own line, says where modality advice came from (endpoint listing / knowledge base / name heuristic — the last one explicitly flagged low-confidence), and renders reference capacities (context window, max output) in a separate read-only block marked "hints only, never auto-filled". Values are thousands-grouped so you can copy them straight into the official capacity inputs by hand.
Auto-adapt: a built-in model knowledge base (DeepSeek V3/V4/R1 with its vision experiment — re-checked 2026-09, when the current official ids became deepseek-flash = V4.1-Flash and deepseek-v4-pro = V4-Pro-0813, the older v4 spellings being compatibility aliases; the official enumeration is Off / low / high / max with high as default; plus dedicated entries for GPT-6 Astra — which rejects none with a 400 — and GPT-5.6-cyber; OpenAI GPT-4o/GPT-4.1/GPT-5.1–5.6 by generation including the codex variants, the o-series, the gpt-oss open weights and the non-reasoning -chat lines; Claude 3.x/4.5–5 with per-generation effort ladders (only the models Anthropic officially lists as effort-capable get one), Gemini, Grok 4.3–4.6, Mistral Small 2603 / Medium 3-5 (the reasoning_effort models; the deprecated magistral line declares no effort control), Qwen incl. Qwen-VL/QvQ and the 3.8 generation, GLM incl. GLM-4V/4.5V/4.6V/5V and GLM-5.2/5.3, Kimi K2.5/K2.6/K2.7-Code/K3, MiniMax M3's thinking toggle, Doubao, Hunyuan hy3, Step incl. 3.5/3.6/3.7, Baidu ERNIE (no effort control on the official surface) — every entry re-verified against each vendor's official docs in 2026-08 and cross-checked against the public OpenRouter catalog; vision-capable variants carry their own entries so the base stem never claims images for them) plus protocol inference keyed by pi-ai's real wire protocols (openai-completions / openai-responses / anthropic-messages, plus a DeepSeek endpoint dialect from baseURL — only api.deepseek.com, the verified official host) fills recommended levels and wire spellings in one click. Families whose endpoints expose no effort-style control reachable here (Llama, Nova, Phi, Cohere, Perplexity sonar) deliberately carry no entry — the low-confidence generic suggestion is more honest. Compat suggestions are gated per protocol: the openai-completions gate takes thinkingFormat/supportsReasoningEffort, and adaptive-thinking Claude families on anthropic-messages routes get the forceAdaptiveThinking pin that makes pi-ai dispatch their declared efforts as output_config.effort.
Endpoint evidence: Auto-adapt also probes the provider's RAW /models listing through a same-origin host route (credential resolved server-side, never echoed) and fuses the signal by confidence — an explicit "does not reason" wins outright; knowledge-base wire values stay authoritative; every suggestion is labeled high / medium / low so you know what to double-check. The same probe reads modality disclosures (OpenRouter-style architecture.input_modalities, models.dev-style nesting, supported_features/capabilities vision flags, supports_vision) and the advertised context length; an explicit listing outranks the knowledge base, silence changes nothing. The probe mirrors the harness's own model discovery, unchanged from the 0.1.2-rc.1 kernel through 0.1.6-alpha.2: it interrogates the same protocol set (OpenAI-compatible and, newly, Anthropic Messages — its native /v1/models route with x-api-key plus the fixed anthropic-version), accepts the enriched models-map listing shape alongside the standard data array, carries the provider profile's configured request headers (a resolved credential still wins its name), and applies the same 4 MB listing ceiling — deployments that authenticate through a custom header now probe as cleanly as they list. The harness's own attribution headers are deliberately not sent: this is a same-origin diagnostic, not a harness request.
Auto-fill (kept clear of the editing window): at boot, models without a reasoningEfforts declaration get a recommended one — and missing input-modality declarations are filled too (opt out via modalityAutofill: false; declared parts, explicit false, and deliberately unset markers are never touched, and capacities are never written at all). Models added during a session are filled by the browser half, and only once you have left the editing card; the write is optimistic-locked, so it never fights you for the write.
Three intents: all levels off = unset the declaration (back to inheritance — persisted as a reasoningEffortsUnset marker so auto-fill respects it, even across restarts); only off armed = disable reasoning (false); levels armed = write the declaration. The editor stays in sync with official-page re-renders and pushed settings changes without clobbering your in-flight edits.
Composer reasoning-effort slider (full popover replication): when the official model menu (the bottom-right seat's popover) opens, its body is replaced on the same painted frame by the upstream design — the slider (white round thumb, gradient pill track, radiation canvas + flare; levels from the current model's adapter-advertised ladder) with 14px padding, a separator, and ONE model row reading name · current effort › whose click opens the official model list. The official "Effort" drill-in row is gone because the slider IS the effort control; the official menu shell and the bottom-right trigger stay untouched. Dragging commits through the official session model-selection seam (optimistic, rolled back on refusal); a refused selection announces in the menu. Models with fewer than two levels show the quiet hint plus the model row. The replica mounts synchronously with the menu, so no official window flashes first. Model switches keep your level: a switch submitted without an explicit effort re-applies the level you picked in this session — which outranks everything for the session's life — else the model's configured default effort, else the level you last picked for that model (remembered per provider/model id), else the vendor's documented default from the knowledge base, in the same atomic commit so no "Default" state flashes in between (gated by the slider toggle; a model without the level on its ladder stays on the official default). Brand-new sessions and session restores land on the same chain through the projection watcher, wired from the session's birth — not from the first time the model menu opens.
Composer model search (unconditional): while the official model menu shows its model list, a search box is injected above the list. It filters the official rows by provider name, model name and model id (space-separated tokens, case-insensitive), hides provider groups with no match, and shows an empty state when nothing matches. ↓ from the input jumps to the first match, Esc clears the query, and while a query is active the arrow keys move between the visible rows only. It sits above the official list but below the official load notices, and it never touches the menu's own size or scrolling. Unlike the slider it is not behind the settings toggle: it is injected whenever the plugin is active.
Per-model default effort (issue #4): each model row's editor gains a "Default effort" picker — the level every new session starts this model at. It is stored on the model row in the settings document (so it survives restarts and follows the deployment, not the browser) and outranks the remembered levels across sessions; within a session your hand always wins — a level you picked (or an explicit "follow the provider default") is never overridden by any automatic mechanism. The picker lists exactly the model's own declared levels; clearing it restores the memory chain, and an empty pick is simply absent from the document (no marker needed — nothing auto-fills it).
Models-page toggle: the "Reasoning effort slider" switch moved out of the general settings and onto the Models settings page, below the Add provider / Add custom provider actions, inside a boxed container (same item form as the upstream plugin). The toggle rides the official settings.models.footer slot, which it takes unconditionally.
Defensive injection: the injector keys off the official page's DOM (aria-labels / classes). If an official upgrade changes the structure, injection simply stops and the official page is untouched; the next scan re-injects once the structure is back.
Bilingual copy (中文 / English).
Anthropic Messages
/v1/models
x-api-key
anthropic-version
data
models
*-vl*
*vision*
gpt-4o
Self-hosted relays: auto-fill and Auto-adapt pin supportsDeveloperRole: false on openai-completions routes no official host claims, so the system prompt keeps the system role (some upstreams reject developer with 角色信息不正确). Explicit values are never overwritten — the one exception being the endpoint-compatibility pickers: choosing "Unset" for one and saving the card is how you revoke it, and the editor only ever deletes a field it showed. Uncheck every level + Save clears a declaration back to bare provider-default requests, which is the compatibility mode for relays that reject thinking parameters.
Forced-thinking models (ladders without off, e.g. GLM-5.3): provider tests and Default calls would otherwise send thinking: disabled and fail (e.g. 1210) — the host maps them to the ladder's vendor default instead. Set defaultGuard: false to restore the old behavior.
