Kilo Code loads AGENTS.md (or AGENT.md) from the project root automatically, plus subdirectory AGENTS.md files when the agent works there. Other rules are files listed under instructions in kilo.jsonc (project, or .kilo/kilo.jsonc) or ~/.config/kilo/kilo.jsonc (global); legacy .kilocode/rules/ still works. MCP servers go under the mcp key of the same file, and permissions are allow / ask / deny rules under permission. You can connect a ChatGPT Plus/Pro subscription for OpenAI models.
Kilo Code at a glance
| Global instructions | ✓ Yesinstructions in ~/.config/kilo/kilo.jsonc; ~/.kilo/rules/*.md |
|---|---|
| Project instructions | ✓ YesAGENTS.md / AGENT.md; files under instructions in kilo.jsonc or .kilo/kilo.jsonc (e.g. .kilo/rules/*.md); legacy .kilocode/rules/ |
| Nested / subfolder | ✓ Yes Subdirectory AGENTS.md loads when the agent accesses files there |
| AGENTS.md | ✓ Yes Loaded automatically; replaces the deprecated memory bank |
| Skills (SKILL.md) | ✓ Yes.kilo/skills/, ~/.kilo/skills/, .agents/skills/, ~/.agents/skills/; .claude/skills/ with Claude Code compatibility on |
| MCP config | ✓ Yesmcp key in kilo.jsonc / .kilo/kilo.jsonc or ~/.config/kilo/kilo.jsonc |
| Permissions / allowlist | ✓ Yespermission in kilo.jsonc: allow / ask / deny per tool (read, edit, bash, webfetch…) with patterns |
| Sandbox | ✓ Yes Optional OS sandbox (off by default; macOS/Linux, not Windows): limits writes and blocks network |
| Hooks | ◐ Partial Plugins ( .kilo/plugin/*.ts) with events such as tool.execute.before |
| Claude Pro/Max login | ? Not documented Not documented |
| ChatGPT Plus/Pro login | ✓ Yes Connect a ChatGPT Plus/Pro subscription for supported OpenAI models |
| API key / own plan | ✓ Yes Kilo Gateway credits or your own provider keys |
Exact file paths
| Path | Scope | What it does |
|---|---|---|
AGENTS.md | Project | Primary instructions (root + subdirectories) |
kilo.jsonc or .kilo/kilo.jsonc | Project | instructions, mcp, permission |
~/.config/kilo/kilo.jsonc | User | Global config |
.kilo/rules/*.md | Project | Rule files referenced from instructions |
.kilocode/rules/ | Project (legacy) | Still read |
.kilo/skills/<name>/SKILL.md | Project | Skills (also .agents/skills) |
Using a shared AGENTS.md with Kilo Code
Native: loaded automatically from the project root and subfolders. See one AGENTS.md for every agent, or generate one with the AGENTS.md generator.
Copy-paste examples
Rules and MCP in one file
File: kilo.jsonc
{
"instructions": [".kilo/rules/*.md"],
"mcp": {
"context7": {
"type": "local",
"command": ["npx", "-y", "@upstash/context7-mcp"],
"enabled": true
}
}
}
Safe permissions preset
Safe Kilo permissions
File: kilo.jsonc
{
"permission": {
"read": "allow",
"glob": "allow",
"grep": "allow",
"edit": "ask",
"webfetch": "ask",
"bash": "ask"
}
}
This is the documented safe baseline: every edit and shell command asks first, and "Approve always" saves narrow allow rules as you go. Enable the optional sandbox so model-run commands can't write outside the project or reach the network.
Gotchas
- The memory bank feature is deprecated; move it into AGENTS.md.
- Sandboxing is off by default and not available on Windows.
.claude/skillsonly load when Claude Code compatibility is enabled.
FAQ
Does Kilo Code support AGENTS.md?
Where is the Kilo Code MCP config?
Sources
Checked against these official pages on 2026-10-12. Cells marked "Not documented" mean we could not find the behaviour in these docs; it may still exist.