Seb Labs

npx add-mcp 'https://mcp.seblabs.dev/mcp'

Quickstart

1. Get a token. Sign in to the developer console with Google. New accounts receive $5 in credit; no card is required. Use Connect Agent or Create auth token and save it as SEB_API_KEY.
2. Ask for a decision. Send context and 2–20 options.
3. Act on the distribution. Proceed with choice when its probability is high. Hand off or ask for more context when it is not.

curl https://seblabs.dev/api/v1/systemone \
  -H "Authorization: Bearer $SEB_API_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
    "context": "The customer wants to replace a damaged order.",
    "options": ["Returns", "Billing", "Technical support"]
  }'

A live response has this shape. Probabilities sum to one. Values below are examples, not measured results.

{
  "mode": "live",
  "model": "Seb-9B-v2-NVFP4",
  "choice": "Returns",
  "state": 1,
  "options": [
    { "label": "Returns", "probability": 0.94 },
    { "label": "Billing", "probability": 0.04 },
    { "label": "Technical support", "probability": 0.02 }
  ],
  "usage": { "input_tokens": 212, "output_tokens": 0, "cost_nanos": 4452 },
  "latency_ms": 48.3,
  "timings": { "server_ms": 24.1, "e2e_ms": 48.3, "phases": { … } }
}

Options keep your order. state is 4 when the top two probabilities are within 0.1 of each other, a close call worth handing off, and 1 otherwise.

Try a free demonstration

No account yet? Send a preset ID to the same endpoint without a token. The four presets return fixed, illustrative distributions without calling a model.

curl https://seblabs.dev/api/v1/systemone \
  -H 'Content-Type: application/json' \
  -d '{"preset":"enterprise"}'

Preset IDs are enterprise, finance, robotics, and handoff. Responses include "mode": "illustrative". These probabilities explain the interface. They are not model outputs, calibration evidence, or measured latency.

API reference

POST https://seblabs.dev/api/v1/systemone with Authorization: Bearer $SEB_API_KEY. Live inference runs the Seb-9B-v2-NVFP4 model on Seb Labs GPU infrastructure.

Request formats

• Routing: { context, options }. The API maps your option labels to stable keys and returns options with probabilities plus choice.
• Typed questions: { state, questions }. Each question has a type (choice, noul or score), instructions, and criteria. The response contains answers keyed by question.

Limits

• 32 KB request body and 12,000 context characters.
• 2–20 unique options, up to 80 characters each.
• 15-second upstream deadline.
• 120 questions per minute and eight in-flight requests per account.

Errors

• 400: invalid input. 401: missing, invalid or expired token (TOKEN_EXPIRED).
• 503 BACKEND_NOT_CONFIGURED: the requested model is not connected.
• Upstream errors are sanitized. Failed or invalid inference is not charged. API responses are not cached.

Responses include mode: "live", usage, and timings. timings.server_ms is inference-server time. timings.e2e_ms runs from API entry through credit settlement; it excludes your network leg. The playground keeps a pasted token only in the current tab’s memory.

MCP server

The stateless Streamable HTTP endpoint is https://mcp.seblabs.dev/mcp. It uses the official MCP SDK. The free illustrative tool works without a key; live calls require an account token and sufficient credits.

claude mcp add --transport http seblabs https://mcp.seblabs.dev/mcp --header "Authorization: Bearer $SEB_API_KEY"

For stdio clients, use the downloadable connector package. It is distributed directly from this website; no npm registry publication is required.

SEB_API_KEY=your_key npx -y https://seblabs.dev/downloads/seblabs-mcp-1.0.3.tgz

Tools

• seb_demo: inspect one of the four illustrative presets. Free.
• seb_decide: submit a text state and 1–10 typed questions (choice, noul, score).
• seb_route: classify context among 2–20 caller-provided routing options.
• seb_batch_decide: process up to 100 states with shared questions and up to four concurrent requests; each item reports its own error or result.
• seb_fin_analyze: Seb-9B-Fin, in closed beta. Available only to approved beta accounts. Request beta access.

Tools do not execute trades, actuate robots, or call downstream routing targets. The general model is never relabeled as a finance specialist.

Agent plugin

Give your agent a decision skill and the MCP connector together. The skill helps Claude Code and Codex recognize routing, classification, evidence screening, rubric scoring, relevance filtering and batch-labeling tasks. It explains when Seb fits, how to call it, how to use its probabilities, and when deterministic code, retrieval or deeper reasoning fits better.

Create a named token in the developer console and provide it as SEB_API_KEY through your client’s environment or secret manager. Node.js 22+ and npm/npx are required. Keep the token out of prompts and committed files. The plugin’s connector sends inference directly to https://mcp.seblabs.dev/mcp. Install the plugin or configure a standalone MCP entry, so you have one connection.

Claude Code

claude plugin marketplace add https://seblabs.dev/plugins/marketplace.json
claude plugin install seblabs@seblabs

The hosted marketplace verifies the plugin archive’s SHA-256. Restart Claude Code with your token available. Use /seblabs:seb-judgment explicitly or let the agent select the skill when appropriate.

Codex

Download the agent plugin bundle and extract it into a permanent directory. Then register and install it:

codex plugin marketplace add /absolute/path/to/seblabs-agents
codex plugin add seblabs@seblabs

Restart Codex with SEB_API_KEY available. The plugin explicitly forwards that environment variable to its connector. Invoke $seb-judgment or let Codex select it for matching tasks. For another skill-enabled client, copy the bundle’s plugins/seblabs/skills/seb-judgment directory into its supported skills location and configure MCP separately.

When to use Seb

• Route or classify: tickets, intents, topics, sentiment, tool choices and escalation labels → seb_route.
• Judge evidence or a rubric: claim support, binary screens, relevance, urgency or quality levels → seb_decide.
• Label many independent items: shared questions over a collection of text or JSON states → seb_batch_decide.
• Check setup free: use seb_demo; it is illustrative and does not test paid inference.

Retrieve the free seb-decision-guide MCP prompt for full descriptions and valid examples, or read the public agent guide. Live calls spend account credits. Probabilities are model estimates; follow your application’s uncertainty policy and retain human or deterministic controls for consequential actions. The public deployment has no dedicated finance specialist. Automatic tool selection remains the host agent’s choice. This package is directly distributed by Seb Labs and has not been approved in a public plugin directory.

Self-host the open weights

Use the model’s maintained instructions for serving with vLLM, the System One adapter, or Apple MLX. Model licensing and hardware requirements are described in the model card.

Seb-9B-v2 model card and serving guide

hf download ironbcc/Seb-9B-v2 --local-dir seb-9b-v2

# After starting vLLM as described in the model card:
python serve/system1_vllm_backend.py \
  --vllm-url http://127.0.0.1:8000 \
  --served-model seb \
  --tokenizer ./seb-9b-v2 \
  --port 8020 --model-name seb-9b-v2

The loopback addresses above are local serving examples. Put an authenticated HTTPS gateway in front of the adapter before exposing it.

Billing & limits

Pricing

• $0.021 per million input tokens. Output is free. This is half the official Jev rate of $0.042/M, checked October 23, 2026.
• We meter model-tokenized prompts, including template, state, instructions and criteria. The current runtime evaluates questions separately, so shared state is counted for each question.
• Responses include usage.input_tokens and integer usage.cost_nanos (one billion nanodollars = $1). Failed or invalid inference is not charged.

Credits and plans

• $5 starter credit once per Google identity. No card required.
• Credit packs: $10, $25, $50 or $100.
• Optional monthly plans: $12 for $10/month, $30 for $25/month, or $125 for $100/month. Credits roll over. Cancel future renewal in Manage billing.
• Refunds and disputes revoke the corresponding credits. Purchases and subscriptions use Stripe-hosted Checkout, which shows the amount and renewal terms before payment.

Automatic top-ups

Connect a card through Stripe, then explicitly enable automatic top-ups. Set a balance threshold, purchase amount, and monthly automatic-charge cap. Saving a card alone does not enable them.

• Checks run on API use or when settings are saved. A low balance can trigger a charge immediately after enabling.
• Ambiguous payments retain one idempotent attempt. Charges requiring authentication pause top-ups.
• Manual purchases and subscriptions are outside the automatic-charge cap.

Tokens

Use Connect Agent to create a token and copy exact MCP connection instructions, including the secret header, Claude Code command, and Cursor configuration. Or create named tokens manually. Tokens expire after 15, 30 or 90 days, or never. They share the account balance, are shown once, and can be revoked independently. They cannot access billing or create other tokens. Keys are stored only as hashes.

Usage dashboard

The console refreshes usage every five seconds while visible, with 7/30-day charts, per-agent totals, error rates and the latest 20 request records. Recent requests show server inference time and API E2E time. Daily detail begins when tracking is enabled; lifetime totals include earlier usage. No prompts are stored.

Enterprise

Quant VPC and custom weights require scoped onboarding. Scope and delivery are agreed before billing starts.

Benchmarks and limitations

The site presents the October 2026 research snapshot supplied with the Seb Labs handoff. It does not rerun these benchmarks or independently certify their results. The original pitch deck and design specification are preserved.

• Decision Index 0.3: 37 suites and 110,520 requests, with Seb-9B-v2 at 59.16 overall. Clef 27B scores 61.71 overall, while Seb leads or ties the selected four operational areas. The deck’s “all areas” language is narrower than the full index.
• The comparison includes Jev 1.13 at 57.96. Jev 1.14 is reported at 60.70; the site does not imply a win against that later version.
• BF16 latency: 21.7 ms p50 and 36.1 ms p95. NVFP4: 21.9 ms p50, 25.2 ms p95, and 244 batched decisions/s. The handoff sometimes labels 21.7 ms as NVFP4; the pitch’s explicit precision breakdown is used here.
• Latency comparisons use different hardware, API environments, and workload settings. The impulse race is an arithmetic illustration, not a live or same-hardware benchmark. Throughput and serial latency are separate measurements.
• Robotics results are simulation decisions: 200/200 PyBullet and 221/228 MuJoCo. The demos use kinematic grasping. Neither these figures nor a control-frequency diagram establish physical safety.
• The trillion-decisions/day and under-10-ms statements are five-year ambitions. Custom deployments require scoped onboarding.

Financial evaluation notes

The deck summarizes 11 wins or ties in 13 locked financial cells, five statistically significant wins, and no significant losses. This combines specialist versions and heterogeneous metrics; it is not a single composite score. The supplied visual and narrative do not provide per-cell confidence intervals for all 13 cells, so the site does not invent a forest plot.

The FOMC narrative reports 0.708 versus 0.532 and a +0.175 unrounded advantage. The displayed table computes +0.176 from rounded values. BigData lists Sonnet at +0.034 in the deck, while the copied raw run lists masked MCC +0.090; these baselines should not be treated as interchangeable.

The Q1 Herculean window reports about −1.7% versus −11.1%. The raw main-window daily-return difference has a confidence interval crossing zero. Post-window and pooled performance differ, so the selected period does not establish reliable trading alpha. C2B’s 300/300 exposure matches establish policy reproduction, not profitability.

Trading and market results are historical research simulations. They are not investment advice, a forecast, or a live trading service.

Brand resources

Kopyttse brand kit PDF · Icon SVG · Icon PNG · Favicon SVG · Favicon PNG