inkcap

Self-hosted LLM workbench

inkcap

Connect your models. Give them tools. Turn good prompts into loops.

A warm, server-driven place to tinker with local and hosted models—without losing a run when the browser goes away.

Log in
An inkcap conversation showing model reasoning, a code answer, token details, tool controls, and the message composer
Reasoning and tools stay close to the conversation—not buried in a dashboard.

From connection to loop

Tinker once. Keep what works.

  1. 01 · Connect

    Choose your models

    Use a local llama-server, an OpenAI-compatible API, or experimental personal ChatGPT/Codex access.

  2. 02 · Explore

    Chat with tools

    Branch prompts, inspect reasoning, and approve MCP calls while the server keeps the run moving.

  3. 03 · Repeat

    Turn it into a loop

    Save a useful prompt as a manual or scheduled automation. Every run remains an inspectable chat.

Server-owned durable runsPostgreSQL is truthCredentials stay server-sidePlain forms; JavaScript only for chat