Quickstart In development

Launch your first Kova Agent in 5 minutes.

5 minEstimated time

3Install methods

ZeroExternal dependencies

Prerequisites

Docker or Rust 1.93+ (either) · Lurus API KeyAPI KeyA secret credential for accessing the API, much like a password. Each Key can have its own quota and permissions.Learn more → (how to get one) · 8 GB+ RAM (recommended) · basic terminal knowledge. About 5 minutes.

---

Install

Install Kova

Docker (recommended), a prebuilt binary, or build from source — pick whichever you prefer.

Choose an installation method

docker run -d \
  --name kova \
  -p 8080:8080 \
  -v kova-data:/data \
  -e LURUS_API_KEY=sk-your-key \
  ghcr.io/hanmahong5-arch/kova:latest

Download prebuilt binaries from GitHub Releases:

Platform	File
Linux x86_64	kova-linux-amd64
macOS (Apple Silicon)	kova-darwin-arm64
macOS (Intel)	kova-darwin-amd64
Windows	kova-windows-amd64.exe

Rust version requirement

Kova requires Rust 1.93+ (Edition 2024). We recommend managing your toolchain with rustup.

---

Configure

Configuration

Create a configuration file kova.toml:

[server]
port = 8080
data_dir = "./data"

[llm]
# 通过 Lurus API 接入所有模型
provider = "openai-compatible"
base_url = "https://hub.lurus.cn/v1"
api_key = "sk-your-lurus-key"
default_model = "deepseek-chat"

[wal]
# WAL 持久化配置
enabled = true
sync_mode = "normal"  # "normal" | "full" (每次写入 fsync)

[security]
# 可选：启用 WAL 加密
# encrypt = true
# encrypt_algorithm = "aes-256-gcm"

---

Get started

Launch your first Agent

Use REST, the Rust SDK, or the CLI to create an Agent, send a task, and stream its execution.

Once it’s running, pick the integration method you’re comfortable with to create and run your first Agent (see the API Reference for the full endpoint list).

---

Workflow

Create a workflow

A workflow chains multiple steps into an ordered execution pipeline (steps pass values to each other via template variables — see the example below):

curl -X POST http://localhost:8080/api/v1/workflows \
  -H "Content-Type: application/json" \
  -d '{ "name": "content-pipeline", "steps": [
    { "name": "research", "agent": "researcher", "prompt": "研究主题：{{input.topic}}" },
    { "name": "write", "agent": "writer", "prompt": "基于研究报告撰写博客：\n{{steps.research.output}}" },
    { "name": "review", "agent": "editor", "prompt": "审校并优化：\n{{steps.write.output}}" }
  ] }'

# 触发
curl -X POST http://localhost:8080/api/v1/workflows/content-pipeline/run \
  -H "Content-Type: application/json" -d '{"input": {"topic": "边缘计算与 AI 推理"}}'

---

Persistence

Verify persistence

Kill the process, and after a restart the task automatically resumes from the WAL checkpoint.

Test Kova’s crash-recovery capability:

1. Start a long-running task

   kova agent run researcher "写一份 5000 字的深度报告"

2. While the task is running, forcibly terminate the process

   kill -9 $(pgrep kova)

3. Restart Kova

   ./kova serve

4. Check the task status — it resumes automatically from the checkpoint

   kova task status

Why recovery works

Every step writes to the WALWALWrite-Ahead Log. Every state change is written to the log before it executes, ensuring recovery from the log after a crash.Learn more → (with a CRC32 checksum) before it executes. When the process crashes, any step that wasn’t confirmed complete is replayed from the checkpoint after restart — no re-calling the LLM, no lost progress. See Core Concepts for details.

---

Next steps

Next steps

Core Concepts — a deep dive into the Agent, Workflow, and WAL architectureAPI Reference — the complete list of REST API endpointsLurus API — learn about the underlying LLM gateway