One field — paste a link or drop a screenshot right into it. We'll route, score it, and keep it in your history. No login required.
Learn product taste
A public preview of representative corpus examples, edited for clarity without changing the underlying taste call. The full corpus stays private; this page exists to build trust and alignment.
Prompt must stay focused on landing page evaluation. Injection and off-topic requests are blocked.
Fast uses Haiku; switch to Sonnet or Opus for a higher-capability side-by-side.
Leave blank to use the selected tier's OpenRouter default.
Manage your profile, subscription, API access, and integrations.
Keeps a record of the recap and raw input you shared.
Create personal API tokens to call the taste model from your own tools.
Bring tastellm into the tools you already build with — as a quality "booster" that reviews plans, prompts, diffs, and UI without leaving your workflow. Connect any MCP-capable tool to the tastellm server below.
Tune Claude Code, Codex, Cursor, and other software-building agents with your tastellm product taste. The MCP steers the next build step, reviews plans and diffs, verifies the result, and captures your corrections for reviewed learning.
Create a personal API token, then connect your harness to the remote MCP (Model Context Protocol) URL below. The token identifies whose taste preferences apply and whose account receives usage.
Expand your tool for its quick MCP config and deeper install. Claude Code gets a plugin with a skill, reviewer, command, and hooks; Cursor gets an always-on rule plus skill; Codex gets AGENTS.md guidance plus a skill.
Once connected, prompt your tool to use tastellm at each stage:
taste_steer_agent before editing, inject its steering prompt, and verify its acceptance criteria.”taste_capture_build_feedback so it enters the reviewed learning queue.”Every successful remote MCP taste_* tool call counts as one metered Developer API request on the account that owns the supplied ptk_… token. Connection handshakes and tool discovery are not billed. Your usage appears with API usage on your account.
Developer API
Send a URL, screenshot, Figma frame, GitHub link, app-flow instructions, or a plain product question. Optional parameters tell tastellm what evidence to use, what audience to judge for, and how detailed the response should be.
API parameters do three jobs: choose the evidence source, ground the model with product context, and control output/cost.
target_context tells tastellm the product, audience, motion, and success moment.category narrows corpus retrieval for recommendations.audience_tags changes the evaluation lens for URL reviews.use_llm=false gives deterministic URL scoring without narrative model feedback.Developer API responses wrap the taste result with a usage receipt for billing and auditability.
{
"result": { "...": "taste output" },
"usage": {
"request_id": "...",
"billed": true,
"units": 1,
"event_type": "api_usage"
}
}Send an Idempotency-Key on billable POSTs. Repeated calls with the same key replay the original response and avoid double billing.
Idempotency-Key: homepage-eval-2026-06-18POST /v1/api/recommend
Use for product strategy or UX questions when there is no artifact to inspect. Shape with query, product_context, category, and output_format.
POST /v1/api/evaluate/url
Use for landing pages and public product pages. Shape with target_context, audience_tags, and use_llm.
POST /v1/api/evaluate/image
Use when a page is private, blocked, or easiest to judge visually. Upload an image and pass target_context.
POST /v1/api/evaluate/figma
Use Figma file/frame URLs. Include ?node-id= to focus one frame and add target_context.
POST /v1/api/evaluate/github
Use repository, pull request, branch, or file URLs. target_context tells tastellm which product surface matters.
POST /v1/api/capture/app-flow
Use for authenticated screens. Shape with public_context_urls, flow_instructions, focus_areas, and screenshot limits.
POST /v1/api/compare
See what tastellm adds over raw model feedback. Shape with prompt, compare_model_tier, and openrouter_model.
POST /v1/api/evaluations
Create URL, Figma, or GitHub jobs and poll GET /v1/api/evaluations/{id}, optionally with webhook_url.
This returns score, grade, dimension scores, recommendations, corpus citations, an agent prompt, and a usage receipt.
curl -X POST https://tastellm.com/v1/api/evaluate/url \
-H "Authorization: Bearer ptk_YOUR_TOKEN" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: homepage-eval-v1" \
-d '{
"url": "https://example.com",
"target_context": "AI-native accounting product for startup founders. Primary conversion is booking a demo.",
"audience_tags": ["startup-founders", "finance", "b2b-saas"],
"use_llm": true
}'overall_scoreNumeric quality signal for evaluations.gradeLetter grade for quick UI summaries.dimension_scoresMachine-readable areas, observations, and suggestions.recommendationsConcrete changes to make.citationsCorpus grounding behind the advice.agent_promptCopyable prompt for a coding agent.moderag or finetuned for recommendations.usageBilling receipt and request id.Pricing
Run product evaluations in the app, bring tastellm into your coding tools through MCP, and use the metered API when your own agents need product judgment on demand.
$0
For trying the evaluator and seeing how corpus-grounded taste feedback feels.
$200/mo
For builders who want repeat evaluations, provider comparison, and account-backed history.
$0.05/call
For agentic development systems that need product judgment programmatically.
Use tastellm across the product surfaces a customer actually experiences.
Connect your own agents to a product-taste check before, during, and after they build.
The product is built around a reviewed corpus and customer-specific taste preferences.
Paste their landing page URL, run an evaluation, then copy the coding-agent prompt and show how the critique turns into buildable fixes.
Connect the MCP server, ask the coding tool to review a plan or diff with tastellm, then apply the top recommendations.
Create a `ptk_` token, call `/v1/api/evaluate/url`, and show the result plus usage receipt, structured error model, and idempotency key support.
Free trial evaluation, Pro at $200/month with 100 included evaluations, and developer API usage at $0.05 per call after 25 free monthly calls.
Service Status
Check whether tastellm is operating normally, see core component health, and read public updates during deployments or outages.
Important deploy, outage, and recovery notes will stay here for customers and builders.
Paste a screenshot, describe what design taste to capture, and we'll format it for admin review before it enters the core model.
Review taste insight contributions and evaluation feedback before they enter the corpus.