dsh-mistral-vibe-provider
Manifest valid★ 3Mistral Vibe Code as a DSH provider route — PKCE browser sign-in, the live model catalogue with per-model prices, the Vibe Code and Studio/API budget meters in Settings and the sidebar, and a per-sess
A DeepSeek Harness provider plugin for Mistral's Vibe Code subscription.
PKCE browser sign-in · live model catalogue · real allowance meters · per-Session cost.
Overview · Features · Requirements · Installation · Configuration · Images · Routes · Troubleshooting · Development
Overview
This plugin registers the LLM provider route mistral-vibe against
https://api.mistral.ai/v1 and adds the account surface around it: sign-in
through Mistral's own console flow, the two allowance meters, prices on every
catalogue model, and a cost figure for the Session in progress.
It is a community integration. It is not affiliated with Mistral, and usage made through the route is billed by Mistral against the same allowance the Vibe CLI draws on, under your own plan and terms.
Features
Sign-in. Settings → Mistral Vibe → Sign in with Mistral runs
console.mistral.ai/api/vibe/sign-in, opens the approval page, polls the process,
and stores the minted key in the credential store. No copy-paste, and no key value
in the profile YAML.
Two allowance meters. GET /v1/billing/budget reports the Vibe Code allowance
and the Studio/API allowance separately. Both are drawn on the account card with
percentage, used-of-total, currency and reset date, and again in the sidebar foot
above the account row, where a click re-reads them and a ten-minute timer keeps
them current. GET /v1/billing/subscription supplies the plan name.
A live catalogue with prices. /v1/models fills the model picker. Every
catalogue model carries its own price line, for example
$0.30 in · $0.007 cached · $1.20 out / 1M tokens for DeepSeek V4.1 Flash 1017,
taken from Mistral's model card. Rows the key cannot use stay visible but marked,
so a tier_not_allowed rejection is a badge rather than a mid-turn surprise.
Cost, per model and per Session. The picker badges each model with its
input/output price ($0.3/$1.2). The composer's token-usage pill gains
· ≈$1.51 for the Session in progress, folded from that Session's own
assistant/message usage at list prices — so a model switch mid-Session prices
each message on the model that actually ran it. The Settings page adds a Usage
panel with token, cache and cost totals across recent Sessions and the most
expensive Sessions by cost.
Thinking levels. Each model's reasoning levels are mapped onto
reasoning_effort, and reasoning is streamed as a thinking block rather than as
answer text. The default level is high for every model that can think, matching
the official DeepSeek route.
Managed images. Screenshots are downscaled, capped at Mistral's 20-image ceiling through the harness's durable offload protocol, and kept visible by default so a visual workflow can compare one render against the next. See Images.
Requirements
| DeepSeek Harness | any 0.1.x or 0.2.x host that mounts dsh-compaction-image-offload (the shipped base does) |
| Node.js | 20 or newer |
| Mistral account | a Vibe Code subscription, or a Studio/API key with access to the models you pick |
Installation
The package is not published on npm yet, so install it from the repository.
dsh plugin forwards its arguments to pnpm inside the profile directory, which
means any spec pnpm accepts works: a path, file:, github:owner/repo, or a
published version.
From the repository:
dsh plugin --profile <profile> add github:Khuraie/dsh-mistral-vibe-provider
From a clone — the development setup, because the profile links the checkout and an edit is live after a restart:
git clone https://github.com/Khuraie/dsh-mistral-vibe-provider
dsh plugin --profile <profile> add ./dsh-mistral-vibe-provider
dsh plugin installs the dependency only. The launcher stacks the patch layers
listed in dsh.profile.bundles, so the package name has to be registered there
too, in ~/.dsh/profiles/<profile>/package.json:
"dsh": {
"profile": {
"bundles": ["…", "@khuraie/dsh-mistral-vibe-provider"]
}
}
The in-app plugin manager performs both steps in one action. Restart DSH afterwards.
Once the package is published, the shortest form becomes
dsh plugin --profile <profile> add @khuraie/dsh-mistral-vibe-provider.
First run
- Open Settings → Mistral Vibe.
- Sign in with Mistral, or paste a Vibe Code API key and select Save key.
- Select Fetch available models to persist the live catalogue.
- Select Refresh on the account card to read the meters.
Configuration
All fields live in the plugin's row in the profile patch. Secret values are references, never literals.
| Field | Default | Purpose |
|---|---|---|
displayName | Mistral Vibe | Name shown in the picker and settings |
baseURL | https://api.mistral.ai/v1 | API base; the affinity header is only sent to *.mistral.ai |
apiKeyEnv | MISTRAL_API_KEY | Credential reference holding the route key |
apiKey | — | Route key stored in the row; prefer the credential store |
studioApiKey | — | Studio/API key for account reads; falls back to the route key |
disabledModels | [] | Model ids hidden from the picker |
bypassRewrite | true | Send x-bili-passthrough: 1 so an in-process rewriter relays the body verbatim |
imageMaxEdge | 1024 | Longest edge, in pixels, of a request image |
imageMaxBytesKb | 768 | Encoded byte budget for one request image |
offloadSeenImages | false | Opt-in cache mode; see Images |
imageKeepBatches | 3 | In that mode, how many recent image batches keep their pixels |
sendAffinity | true | Send x-affinity so a Session stays on one backend instance |
allowRemote | true | Answer this page when the web UI is exposed over HTTP/S |
accessToken | — | Token required for non-same-origin requests |
discoveredModels | [] | Ids a live read reported |
servedModels | [] | The whole served surface of the last live read |
modelsJson | — | The model list a previous fetch stored |
discoveredInfo | — | Cached metadata for discovered rows |
Images
Mistral's prompt cache is a longest-common-prefix cache: it serves a request only when it repeats the previous one byte for byte. An image payload is the largest thing that can change, so this route manages images explicitly instead of forwarding whatever the harness hands it.
Downscaling
Every image is re-encoded before it is sent: 1024 px on the longest edge and
768 KB encoded. In a measured Session, steps carrying a 1–4 MB screenshot fell
to 1–26% cache hit while text-only steps in the same Session stayed at 86–100%.
The harness default for other providers is 1568 px / 4 MB; this route trades that
detail for cache. Raise imageMaxEdge or imageMaxBytesKb for finer detail at
the cost of cache hits.
The 20-image ceiling
Mistral rejects a request carrying more than 20 images with
400 invalid_request_invalid_args ("Total number of images exceeds the maximum
allowed of 20", code 3051). The route does not drop images itself — that would
rewrite the transcript on every request and break the prefix cache. It fails the
step with IMAGE_OFFLOAD_REQUIRED and the count it needs; the harness's
dsh-compaction-image-offload plugin records one durable image/offload
selection and retries. The count is retired four at a time, so the retained window
stays between 17 and 20 images instead of collapsing to one.
An offloaded image becomes a named placeholder rather than a silent gap:
[image omitted from this request to stay inside the image limit (not a failed
read); "render-04.png" (sha256:…)]
The "not a failed read" wording is deliberate: without it a model finds the
placeholder where its own read_image result used to be and reports the read as
broken, then re-reads the file.
Keeping images visible
Images stay on the wire up to the ceiling by default, because a visual workflow — render, look, edit, re-render — needs the previous render to compare against.
offloadSeenImages is the opt-in cache mode: once the model has answered on an
image, that image becomes a placeholder, so a new screenshot no longer rolls the
cache back. The cost is that seeing an old image again means reading it again.
imageKeepBatches decides how many recent batches stay as pixels in that mode.
Measured
| Session | Images | Offloads | Cache efficiency |
|---|---|---|---|
| 52-image judging, earlier batch-cap design | 52 | 4 | 78.8% |
| 192-step visual iteration, current default | 40 | 17 | 95.4% |
Keeping images and letting the budget retire four at a time measured better than the aggressive offloading it replaced: fewer transcript rewrites means more of the prefix survives.
Routes
| Route | Effect |
|---|---|
GET/POST /mistral-vibe/config | The row as the page needs it; no key values |
POST /mistral-vibe/discover | Merge a live /v1/models read into the stored list |
GET /mistral-vibe/budget | Plan name plus the Vibe and Studio/API meters |
GET /mistral-vibe/session-cost?sessionId=… | What one Session's usage cost at list prices |
GET /mistral-vibe/usage | Token, cache and cost totals across recent Sessions |
POST /mistral-vibe/sign-in | Start a PKCE attempt and answer the console URL |
GET /mistral-vibe/sign-in/status?processId=… | Poll an attempt; on approval, store the key |
Every route passes DSH's own connection.requestRejection check before its
handler runs. A route mounted with webServer.register gets none of the host's
Host/Origin admission for free, and without that check a page on another origin
could POST to these routes. Requests that are not same-origin must present the
configured accessToken; that token is the one documented exception, because
another page cannot read it.
Troubleshooting
| Symptom | Cause and remedy |
|---|---|
401 access_token_required | The request is not same-origin and no accessToken is configured. Set one, or reach the page over loopback. |
403 {"type":"tier_not_allowed","code":"1910"} | The model is not included in your plan. The picker marks rows that never appeared in a live read. |
400 … images exceeds the maximum allowed of 20 | The harness's image-offload plugin is not mounted, so IMAGE_OFFLOAD_REQUIRED reached ordinary recovery. Use a host that ships it. |
| Settings shows no key while chat works | The key is coming from the process environment or a legacy reference. The page reads the same lookup the route does; if they disagree, restart the host. |
| The Usage panel stays on Reading… | The first scan prepares up to 100 cold Sessions and takes a few seconds. Later reads are cached for five minutes. |
429 with a reset time | The route records retry-after and refuses further requests until the reset, rather than spending the quota on certain failures. |
Development
node scripts/test-model-pricing.cjs # catalogue, pricing, images, settings projection
node scripts/test-client-surfaces.cjs # slots, sidebar card, composer cost run
node scripts/test-deepseek-readiness.mjs # model readiness and wire body
node scripts/test-tool-images.mjs # tool-result images on the wire
node scripts/test-wire-options.mjs # stop, effort mapping, stored-model defaults
The host half (lib/index.js) is plain ESM with no @deepseek-ai/* imports; the
client half (lib/client.js) is a lazy CommonJS bundle loaded through
window.__ModuleLoader__. docs/DETAIL.md records the design notes behind the
wire adapter, and docs/market-entry.yml is the catalogue submission.
Credits
The wire adapter, catalogue, thinking-level contract and client shell are derived
from dsh-mistral-vibe by
Meltartica (MIT). This bundle replaces its settings half with the account card and
adds the sign-in, billing and usage routes. See NOTICE.
License
MIT © 2026 Meltartica and Mars-Sea — see LICENSE and NOTICE.
Comments
Loading…
Similar plugins
by log-li
Peak / off-peak rate badges for the model selector (every row) and the composer tool row, judged per provider by its own time zone and schedule; the live catalog is served from the host over a fenced
★ 1
↓ 705/wk
MIT
TypeScript
Sep 21, 2026
dsh plugin --profile web add dsh-peakrateby VICqiufeeng
Use the MiniMax Code (Mavis) subscription already signed in on this machine as a DeepSeek Harness model provider — with self-renewing credentials, a points/check-in panel and a model roster editor.
★ 2
MIT
JavaScript
Nov 4, 2026
dsh plugin --profile web add dsh-minimax-connectby Mars-Sea
Command Code provider plugin for DeepSeek Harness (dsh). Adds Command Code model access, live model catalog, plan-aware model selection, reasoning effort, image input, web search, and multi-account su
★ 487
↓ 4.3k/wk
MIT
TypeScript
Nov 18, 2026
dsh plugin --profile web add @mars-sea/dsh-commandcode-providerby honoriomelo
DSH Web GUI plugin: adds a live search field inside the model picker menu of the composer, plus the Effort (reasoning) selector. Drop-in replacement for the native model seat, sharing the same per-ses
★ 0
MIT
JavaScript
Oct 31, 2026
dsh plugin --profile web add dsh-model-picker-searchOpenRouter provider for DeepSeek Harness with a live model list, prices and spend tracking
★ 0
MIT
JavaScript
dsh plugin --profile web add dsh-openrouter-liveby itsbuyru-bit
Per-model API price table with Peak and Off-Peak rates in USD or CNY, the DeepSeek account balance and live session spend under the composer, an all-time total, a collapsible source URL and check date
★ 0
MIT
JavaScript
Oct 26, 2026
dsh plugin --profile web add dsh-model-price