Repository navigation
Forge 0.3.0
Forge 0.3.0
Forge 0.3.0 is synchronized with Grok Build 0.2.112. This release simplifies Forge's provider architecture while preserving ChatGPT subscription-backed Codex models, provider-aware controls, and external Claude Code and Codex CLI harnesses.
Provider architecture
Generic third-party providers now use upstream [model_providers.*] configuration and credentials. Forge no longer maintains a parallel [provider.*] configuration format, provider key store, status layer, or provider-login interface.
ChatGPT subscription support remains available through a narrow compatibility layer. Forge reads Codex OAuth data from ~/.codex/auth.json only for the canonical ChatGPT Codex endpoint. Fast Mode remains session-scoped and sends service_tier = "priority" only for models that explicitly support it.
OpenRouter and other generic providers use ordinary upstream API-key or environment-key configuration. Forge does not store their credentials.
Endpoint and data isolation
Provider-specific behavior now fails closed:
- xAI bearer tokens and private request metadata are stripped from unknown, cleartext, proxy, and third-party endpoints.
- Codex OAuth-file fallback cannot be activated by merely reusing the expected environment-key name on another endpoint.
- ChatGPT Codex Responses requests always send
store = false. - Cross-provider history drops opaque reasoning and flattens backend-only tool calls while retaining portable conversation context.
Codex compatibility
- Streamed reasoning summaries preserve indexed part boundaries, preventing adjacent Markdown sections from rendering as glued text.
- Credential reload resolves the complete active model, including provider defaults and model overrides.
- Provider-aware
/usage, reasoning-effort controls, and Claude Code/Codex CLI harness integrations remain supported.
Upstream synchronization
This release includes Grok Build 0.2.112 improvements such as faster nonblocking startup, bounded memory during large session replay, corrected plan-mode ordering, improved /loop termination prompts, configurable subagent nesting, leader-process sandboxing, and cleaner background-task snapshots.
Migration
Replace legacy Forge provider configuration with upstream provider entries:
[model_providers.openrouter]
base_url = "https://openrouter.ai/api/v1"
env_key = "OPENROUTER_API_KEY"Model entries select the provider with model_provider. Run codex login to populate or refresh ChatGPT Codex OAuth credentials.
Supported platforms
- macOS Apple Silicon (
aarch64-apple-darwin) - Linux x86_64 (
x86_64-unknown-linux-gnu) - Linux AArch64 (
aarch64-unknown-linux-gnu)
Install or update
curl -fsSL https://raw.githubusercontent.com/DeveshParagiri/forge/main/scripts/install | shExisting installations can update after release artifacts finish publishing:
grok updateChecksums are published beside every archive. See CHANGELOG.md for the complete Forge release history.