Kavara Kirk

Structural change detection on raw, unlabeled data. CPU-only. No training.

Kavara Kirk is the first algorithm in the Ulysses family — an algorithmic primitive for detecting structural change in raw, unlabeled data streams. Think of it as a Kalman filter for the non-Gaussian, non-stationary world.

MCP connector

Server URL: https://kirk-mcp.kavara.ai/mcp
Login: None
Protocol: MCP 2025-11-25 · Interactive UI
Instructions: Kirk-mcp — Kavara sealed inference engine, attested with engine_sha on every response.
version: 0.2.0 served_from: kavara-ai/kirk-mcp @ unknown
Kirk is an explicit energy-based inference primitive. Properties: no collapse by construction, online learning without pretraining, entropy differential as native surprise/anomaly signal, and composability across heterogeneous inputs.
KIP API keys use Authorization: Bearer k_... on https://kirk-mcp.kavara.ai/mcp. Legacy personal connector URLs and CF Access service-token credentials remain supported.
Kip account policy
Data fit is free: managed KIP keys spend zero credits while exploring admitted inputs, with no credit-card requirement or fixed evaluation allowance.
Throughput limits apply: two concurrent computations per account, 60 runs/minute, at most 50 books per run and 256 KiB input. These are capacity controls, not a trial expiry.
Ongoing execution requires an explicitly accepted workflow quote, X-Kavara-Workflow and a UUID-v4 Idempotency-Key. Never infer consent from buying credits or repeating a configuration.

Tools

Reads data13

Changes data1

Skills

None in the store listings or on the server

MCP connector

Server URL

https://kirk-mcp.kavara.ai/mcp

None

Protocol

MCP 2025-11-25 · Interactive UI

Instructions

Kirk-mcp — Kavara sealed inference engine, attested with engine_sha on every response.

version: 0.2.0 served_from: kavara-ai/kirk-mcp @ unknown

Kirk is an explicit energy-based inference primitive. Properties: no collapse by construction, online learning without pretraining, entropy differential as native surprise/anomaly signal, and composability across heterogeneous inputs.

KIP API keys use Authorization: Bearer k_... on https://kirk-mcp.kavara.ai/mcp. Legacy personal connector URLs and CF Access service-token credentials remain supported.

Kip account policy

Data fit is free: managed KIP keys spend zero credits while exploring admitted inputs, with no credit-card requirement or fixed evaluation allowance.

Throughput limits apply: two concurrent computations per account, 60 runs/minute, at most 50 books per run and 256 KiB input. These are capacity controls, not a trial expiry.

Ongoing execution requires an explicitly accepted workflow quote, X-Kavara-Workflow and a UUID-v4 Idempotency-Key. Never infer consent from buying credits or repeating a configuration.