Skip to content

rgf-examples.md

Latest commit

 

History

History
368 lines (301 loc) · 9.87 KB

rgf-examples.md

File metadata and controls

368 lines (301 loc) · 9.87 KB

.rgf Configuration Examples

Real-world configuration examples for different deployment scenarios.

Table of Contents

Basic Single-Agent Setup

For a single developer running one agent locally:

[global]
currency = "USD"
fail_safe_action = "deny"

[agents.dev-agent]
daily_budget = 5.00
max_tokens_per_request = 8192
allowed_models = [
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/deepseek/deepseek-v4-flash",
    "openrouter/google/gemini-3-5-flash"
]

Multi-Agent Development Team

Three agents with different roles and budgets:

[global]
currency = "USD"
fail_safe_action = "deny"

[agents.coder]
daily_budget = 8.00
max_tokens_per_request = 16384
allowed_models = [
    "openrouter/anthropic/claude-opus-4-7",
    "openrouter/openai/gpt-5-4",
    "openrouter/x-ai/grok-build-0-1"
]
allowed_tools = ["web_search", "code_exec", "file_read", "git"]

[agents.reviewer]
daily_budget = 3.00
max_tokens_per_request = 8192
allowed_models = [
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/google/gemini-3-5-flash"
]
allowed_tools = ["web_search", "file_read", "git"]

[agents.researcher]
daily_budget = 4.00
max_tokens_per_request = 4096
allowed_models = [
    "openrouter/deepseek/deepseek-v4-flash",
    "openrouter/qwen/qwen3-7-max",
    "openrouter/google/gemini-3-5-flash"
]
allowed_tools = ["web_search", "web_scraper", "file_read"]

CI/CD Pipeline Agents

Automated agents for CI/CD with tight budgets and restricted models:

[global]
currency = "USD"
fail_safe_action = "deny"

[agents.ci-test-runner]
daily_budget = 2.00
max_tokens_per_request = 4096
allowed_models = [
    "openrouter/deepseek/deepseek-v4-flash",
    "openrouter/openai/gpt-5-4-mini"
]
allowed_tools = ["code_exec", "file_read", "git"]

[agents.ci-deployer]
daily_budget = 1.00
max_tokens_per_request = 2048
allowed_models = [
    "openrouter/deepseek/deepseek-v4-flash"
]
allowed_tools = ["code_exec", "git"]

[agents.pr-reviewer]
daily_budget = 3.00
max_tokens_per_request = 8192
allowed_models = [
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/google/gemini-3-1-pro"
]
allowed_tools = ["file_read", "git", "web_search"]

Production with Model Restrictions

Restrict to specific model tiers for cost control:

[global]
currency = "USD"
fail_safe_action = "deny"

[agents.prod-primary]
daily_budget = 20.00
monthly_budget = 400.00
max_tokens_per_request = 16384
allowed_models = [
    # Premium tier — only for complex reasoning
    "openrouter/anthropic/claude-opus-4-7",
    "openrouter/openai/gpt-5-5",
    # Standard tier — daily work
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/google/gemini-3-5-flash",
    # Fast tier — simple tasks
    "openrouter/deepseek/deepseek-v4-flash",
    "openrouter/openai/gpt-5-4-mini"
]
allowed_tools = ["web_search", "code_exec", "file_read", "git", "web_scraper"]

[agents.prod-secondary]
daily_budget = 5.00
monthly_budget = 100.00
max_tokens_per_request = 8192
allowed_models = [
    # No premium tier for secondary agents
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/google/gemini-3-5-flash",
    "openrouter/deepseek/deepseek-v4-flash"
]
allowed_tools = ["web_search", "file_read"]

Budget-Conscious Startup

Maximize free and cheap models:

[global]
currency = "USD"
fail_safe_action = "allow"  # Allow through if budget system fails

[agents.main]
daily_budget = 2.00
max_tokens_per_request = 8192
allowed_models = [
    # Free tier models (OpenRouter :free suffix)
    "openrouter/meta-llama/llama-3-3-70b-instruct:free",
    "openrouter/deepseek/deepseek-r1:free",
    "openrouter/qwen/qwen3-next-80b-a3b-instruct:free",
    "openrouter/google/gemma-4-26b-a4b-it:free",
    # Cheap paid fallback
    "openrouter/deepseek/deepseek-v4-flash",
    "openrouter/openai/gpt-oss-120b:free"
]
allowed_tools = ["web_search", "code_exec", "file_read"]

Enterprise Multi-Tenant

Multiple organizations with isolated budgets:

[global]
currency = "USD"
fail_safe_action = "deny"

# Organization A — Engineering team
[agents.org-a-coder]
daily_budget = 15.00
monthly_budget = 300.00
max_tokens_per_request = 16384
allowed_models = [
    "openrouter/anthropic/claude-opus-4-7",
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/openai/gpt-5-4",
    "openrouter/google/gemini-3-5-flash"
]
allowed_tools = ["web_search", "code_exec", "file_read", "git", "web_scraper"]

[agents.org-a-reviewer]
daily_budget = 5.00
monthly_budget = 100.00
max_tokens_per_request = 8192
allowed_models = [
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/google/gemini-3-1-pro"
]
allowed_tools = ["file_read", "git"]

# Organization B — Marketing team (lower budget)
[agents.org-b-writer]
daily_budget = 3.00
monthly_budget = 60.00
max_tokens_per_request = 4096
allowed_models = [
    "openrouter/anthropic/claude-sonnet-4-6",
    "openrouter/google/gemini-3-5-flash",
    "openrouter/deepseek/deepseek-v4-flash"
]
allowed_tools = ["web_search", "file_read"]

[agents.org-b-analyst]
daily_budget = 2.00
monthly_budget = 40.00
max_tokens_per_request = 4096
allowed_models = [
    "openrouter/deepseek/deepseek-v4-flash",
    "openrouter/google/gemini-3-5-flash"
]
allowed_tools = ["web_search", "web_scraper"]

GitHub Copilot as LLM Provider

Use GitHub Copilot models through ResGov. GitHub Copilot uses a different auth flow (OAuth token via api.githubcopilot.com) and has its own pricing model.

Available GitHub Copilot Models (as of May 2026)

Model Provider Tier Multiplier
GPT-5.5 OpenAI Premium 7.5x
GPT-5.4 OpenAI Standard 1x
GPT-5.4 mini OpenAI Fast 0.33x
GPT-5.3-Codex OpenAI Standard 1x
GPT-5.2 OpenAI Standard 1x
GPT-5 mini OpenAI Free 0x
GPT-4.1 OpenAI Free 0x
Claude Opus 4.7 Anthropic Premium 15x
Claude Opus 4.6 Anthropic Premium 3x
Claude Opus 4.5 Anthropic Premium 3x
Claude Sonnet 4.6 Anthropic Standard 1x
Claude Sonnet 4.5 Anthropic Standard 1x
Claude Haiku 4.5 Anthropic Fast 0.33x
Gemini 3.5 Flash Google Premium 14x
Gemini 3.1 Pro Google Standard 1x
Gemini 3 Flash Google Fast 0.33x
Gemini 2.5 Pro Google Standard 1x

Note: GitHub Copilot is transitioning to usage-based billing (token-based) in June 2026. Model multipliers will be replaced by per-token pricing. See GitHub Copilot Billing.

GitHub Copilot .rgf Configuration

[global]
currency = "USD"
fail_safe_action = "deny"

[agents.copilot-dev]
daily_budget = 10.00
max_tokens_per_request = 16384
# GitHub Copilot model names (as used in api.githubcopilot.com)
allowed_models = [
    "gpt-5-4",
    "gpt-5-4-mini",
    "gpt-5-3-codex",
    "claude-sonnet-4-6",
    "claude-haiku-4-5",
    "gemini-2-5-pro",
    "gemini-3-5-flash"
]
allowed_tools = ["web_search", "code_exec", "file_read", "git"]

GitHub Copilot Auth Setup

GitHub Copilot requires an OAuth token (not an API key). To use it with ResGov:

  1. Generate a GitHub OAuth token with copilot scope
  2. Set it as GITHUB_COPILOT_TOKEN in your .env
  3. ResGov routes requests to https://api.githubcopilot.com/chat/completions
# .env
GITHUB_COPILOT_TOKEN=github_oauth_token_here

GitHub Copilot Pricing Tiers (May 2026)

Plan Price Premium Requests Notes
Free $0 50/mo 2,000 completions + GPT-4.1/GPT-4o unlimited chat
Pro $10/mo 300/mo Full model access
Pro+ $39/mo 1,500/mo Highest tier for individuals
Business $19/user/mo 300/user/mo Org-level management
Enterprise $39/user/mo 1,000/user/mo SSO, audit logs, policy controls

Important: Premium requests are consumed based on model multipliers. A single Claude Opus 4.7 request costs 15 premium requests. GPT-5 mini and GPT-4.1 are free (0 multiplier).

Parameter Reference

Global Parameters

Parameter Type Default Description
currency string "USD" Budget currency for display
fail_safe_action string "deny" Behavior when budget system fails: "deny" (block all) or "allow" (pass through)

Agent Parameters

Parameter Type Default Description
daily_budget float unlimited Maximum daily spend in USD
monthly_budget float unlimited Maximum monthly spend in USD
max_tokens_per_request int unlimited Maximum tokens per LLM request
allowed_models list all models Whitelist of model IDs (OpenRouter or GitHub Copilot format)
allowed_tools list all tools Whitelist of tool names the agent can use

Model ID Formats

OpenRouter format: provider/model-id

allowed_models = [
    "openrouter/anthropic/claude-opus-4-7",
    "openrouter/deepseek/deepseek-v4-flash",
    "openrouter/google/gemini-3-5-flash",
    "openrouter/openai/gpt-5-4",
    "openrouter/qwen/qwen3-7-max",
    "openrouter/x-ai/grok-build-0-1"
]

OpenRouter free tier: Append :free to model ID

allowed_models = [
    "openrouter/deepseek/deepseek-r1:free",
    "openrouter/meta-llama/llama-3-3-70b-instruct:free"
]

GitHub Copilot format: Use model name directly

allowed_models = [
    "gpt-5-4",
    "claude-sonnet-4-6",
    "gemini-2-5-pro"
]