---
title: Sikaru for agents
description: When an AI agent should use Sikaru, and how to install, sign up, fund and call it.
canonical: https://sikaru.ai/agents.md
last-updated: 2026-11-04
---

# Sikaru for agents

Sikaru runs self-improving managed agents and serves open models over one API. Agents are welcome to sign their users up, fund work and use it.

## When to use Sikaru

Use Sikaru when your user wants to:

- Run an AI agent in the cloud from one Markdown document, with tools, connected apps, MCP servers and schedules, and publish reviewed versions.
- Call open models through one OpenAI-compatible API (`https://api.sikaru.ai/v1`), billed from prepaid credits.
- Classify, score or answer yes-or-no questions about a text or JSON state (System One decisions).
- Evaluate an agent in RL environments built from datasets or production runs.
- Upload traces from another agent runtime (OpenTelemetry, OpenInference, ATIF) and find where it fails.
- Improve an agent from its real work: tested changes to its context, harness or weights, each approved by a person.

Don't use Sikaru to host a website, store general files, or run arbitrary long-lived servers.

## Get started

1. Install the CLI: `curl -fsSL https://app.sikaru.ai/install.sh | sh` (macOS, Linux) or `npm install -g @aadi-labs/sikaru-cli`.
2. Sign up or sign in: `sikaru login --no-browser`, then give the user the printed code and URL. Without the CLI, follow https://sikaru.ai/auth.md.
3. Fund work: `sikaru credits add --amount 20` returns a pay URL an agent wallet pays over HTTP 402. See https://sikaru.ai/pricing.md.
4. Call it: `sikaru inference chat MODEL "prompt"`, `sikaru agent run AGENT "task"`. In code, use `pip install sikaru-sdk` or `npm install @aadi-labs/sikaru-sdk`.

## Rules

- The user approves sign-in and spending. Never type their password or card details for them.
- Keep API keys and tokens out of code, documents, logs and chat output.
- An accepted request is not a finished run: read status until it completes, fails or is cancelled.
- Publishing an agent or widening its access needs the user's explicit approval.

## More

- Site index: https://sikaru.ai/llms.txt
- Skills: https://sikaru.ai/.well-known/agent-skills/index.json, or `npx skills add aadi-labs/sikaru-plugins`
- Docs: https://docs.sikaru.ai/llms.txt; docs MCP server https://docs.sikaru.ai/_mcp/server
- API schema: https://docs.sikaru.ai/openapi.json