sber-gigachat_skill

Overview

This skill integrates Sber's GigaChat models with OpenClaw using the gpt2giga proxy. It exposes GigaChat as an OpenAI-compatible endpoint so OpenClaw and agents can use GigaChat-Max, GigaChat-Pro, or GigaChat Lite without manual OAuth handling. The proxy manages tokens and translates request/response formats.

How this skill works

Start a local gpt2giga proxy that reads credentials from environment variables and performs OAuth with Sber automatically. OpenClaw is configured to call the proxy at http://localhost:8443/v1/chat/completions using the OpenAI-completions API format. gpt2giga translates requests to GigaChat format, forwards them to Sber, refreshes tokens as needed, and converts responses back to OpenAI format.

When to use it

• You want to run GigaChat models inside an OpenClaw deployment.
• You need OpenAI API compatibility for agents but prefer GigaChat model flavors (Max/Pro/Lite).
• You need automatic OAuth token handling without embedding secrets in code.
• You want to host a local proxy to control traffic, logging, and retries.
• You need an easy way to add GigaChat agents to existing OpenClaw agent configs.

Best practices

• Store CLIENT_ID and CLIENT_SECRET only in an environment file with strict permissions (chmod 600).
• Run the provided start/stop/status scripts to manage the proxy and keep logs in ~./openclaw for troubleshooting.
• Use the provided patch-config script to safely back up and update openclaw.json rather than editing it manually.
• Rotate models to avoid free-tier quotas: prefer Max → Pro → Lite depending on usage and cost.
• Install Sber root CA to /etc/ssl/certs/sber-ca.crt if you need strict SSL verification; default disables verification.

Example use cases

• Add a Russian-context assistant using GigaChat-Pro for internal support agents.
• Route OpenClaw agent traffic through localhost proxy to test GigaChat-Max capabilities with 32k context windows.
• Run lightweight public-facing assistants on GigaChat Lite to conserve quota and cost.
• Use the proxy in CI or staging to validate integrations without exposing client secrets.
• Recover from token errors by restarting the proxy script rather than handling tokens manually.