KORE

Agent-native ops kernel with 14 PIL-powered tools — route, compress, guard, memory, score, certify, optimizeprompt, and more.

Community: Submitted by a user or imported; check the owner before granting accessOnlineNo sign-inGlobalFreeCan modify data

What it can do

  • Guard: Check AI output against source documents to detect hallucination. Returns safety score, flagged claims, and revision hints.
  • Compress: Semantically compress text 3-5x without quality loss. Reduces token costs for LLM context windows.
  • Route: Route a task to the cheapest LLM provider that meets quality threshold. Uses 11-provider fallback chain. Saves 60-90% on inference costs.

What data it sees

Do you need an account

No: the server works without sign-in

Agent-native ops kernel with 14 PIL-powered tools — route, compress, guard, memory, score, certify, optimize_prompt, and more.

PIPELINE (7cr = $0.007): route → compress → guard → memory → score → certify

PIL KERNEL: optimize_prompt uses a self-improving 45-template prompt archive invent: full invention chain with adversarial verification (15cr)

Free tier: 500 credits. No card needed. POST /v1/register → instant API key

Starter: $9 / 10,000 credits Pro: $49 / 100,000 credits

Server tool list (13)

Raw names from tools/list. Only developers need these.

guardCheck AI output against source documents to detect hallucination. Returns safety score, flagged claims, and revision hints.
compressSemantically compress text 3-5x without quality loss. Reduces token costs for LLM context windows.
routeRoute a task to the cheapest LLM provider that meets quality threshold. Uses 11-provider fallback chain. Saves 60-90% on inference costs.
memoryStore or retrieve observations from cross-agent memory. Qdrant-backed with all-MiniLM embeddings. Network effect across all agents.
normalizeDetect prompt injections, normalize whitespace, strip noise from text. Returns risk score and blocked flag.
scoreRank multiple candidate outputs by heuristic quality. Evaluates length, code quality, structure, and keyword coverage.
splitDecompose a complex task into parallel subtasks. Supports 5 decomposition types: sequential, parallel, build, verify, document.
diffCompute a meaning-preserving semantic diff between original and modified text using SequenceMatcher.
provenanceGenerate EU AI Act compliant provenance certificate with HMAC-SHA256 hash chain. Certify claims with source attribution.
embedCompute text embeddings via all-MiniLM-L6-v2 with LRU cache. Returns vector embeddings for RAG or similarity search.
sandboxExecute Python code in a secure sandbox. 30-second timeout with automatic temp file cleanup. No Docker needed.
verifyVerify AI-generated answers against provided source documents. Returns groundedness score, risk level, and detailed issue list.
auditRetrieve or verify the immutable HMAC-SHA256 chained audit trail. EU AI Act compliant. Tamper-evident logging for all API calls.
KORE: connect to Claude, ChatGPT, Cursor · Connectors.fun