OpenClaw gets a complete Russian-language guide to JSON5 config, CLI, and secure setup

A detailed guide to OpenClaw has been published on Habr — a platform where nearly all operational configuration revolves around a single file, openclaw.json, and an extensive CLI. The material gathers in one place what previously had to be pieced together from the official documentation, GitHub issues, and community posts.

What's in the config

The author breaks down how the top-level configuration is structured and why OpenClaw demands care when editing manually. The file uses JSON5, which allows comments and trailing commas, but the schema is validated very strictly at startup. If an unknown key is added, the gateway simply will not start.

The following sections are described separately: gateway, env, agents, models, auth, channels, session, cron, tools, sandbox, and logging, as well as how hot reload works without a full restart.

• gateway: port, bind mode, access token, and hot reload behavior
• agents and models: base settings, fallbacks, and provider selection
• auth and channels: profile switching, Telegram, Discord, and other channels
• session and cron: dialog isolation, time-based resets, and background jobs
• env and secrets: variable priority, SecretRef, and storing sensitive data

The main value of the material is that it goes beyond a mere list of fields. The text explains how different configuration levels interact, where default values apply, when settings take effect without a restart, and which parameters are best left untouched manually.

For those running OpenClaw not locally but on a VPS or in a production environment, this is no longer just a reference — it is a full operational playbook.

Where configuration breaks

The most useful section is dedicated not to the happy path, but to typical failure modes. The guide emphasizes that some parameters only work in agents.defaults and are silently ignored if moved to an individual agent's settings. This can make the config look correct while the desired behavior never activates.

The recommendation is simple: use the openclaw config get, set, and unset commands more often instead of editing directly, and run diagnostics after every change.

openclaw doctor — the single most useful command in the entire CLI.

The guide also addresses the problem of context bloat. OpenClaw uses two distinct mechanisms: pruning removes old tool results before calling the model, while compaction compresses history when the context window is nearly full. As an example, a case is cited where 35 messages grew into a 2.9 MB session file and a context of 208,467 tokens, causing the system to stop responding when the model's limit was 200,000 tokens.

CLI for admins

The CLI is organized not alphabetically, but by operational scenarios, which makes the guide practical. Commands for starting the gateway, checking daemon status, managing channels, logs, cron jobs, plugins, skills, sessions, and updates are all gathered in one place.

For infrastructure commands, safe modes are highlighted separately: health works even without a running daemon, security audit checks the config for risks, and secrets audit helps catch credential leaks before they reach production.

The final important section covers environment variables and version control. The guide documents the priority order: process environment first, then local .env, then the global .env in the OpenClaw directory, and finally built-in env.vars. For secrets, it is recommended to use variable references or SecretRef rather than values directly in openclaw.json, and to commit only a configuration template to the repository, excluding .env files, session folders, memory files, and temporary state.

What this means

OpenClaw is becoming a tool not only for enthusiasts willing to read source code, but also for teams that need a clear operational playbook. If a project relies on multiple models, channels, and automation, a guide like this reduces the risk of silent failures, secret leaks, and unstable behavior after updates.