ForgeCode
An AI-enhanced terminal development environment — a coding agent with TUI, one-shot CLI, and zsh-plugin modes.
About
ForgeCode (the Forge project from Tailcall) is a terminal coding agent that meets you in three different ways: an interactive TUI, a one-shot CLI mode, and — unusually — a zsh plugin where prefixing commands with : turns your shell history into an agent conversation with file attachments, git integration, and shell-command tools. Provider support is broad (Claude, GPT, O-series, Grok, DeepSeek, Gemini, and 300+ models via configuration), and advanced configuration covers skills, semantic workspace search, and custom agent behavior.
The zsh-plugin mode is the genuinely distinctive idea here: instead of context-switching into a separate agent UI, you stay in the shell you already live in. With ~7,600 stars and same-week commit activity, it is one of the more actively developed terminal agents.
The honest caveat: the "300+ models" claim is provider-configuration breadth, not 300 hand-tuned integrations — quality will vary by model. And three interaction modes means three surfaces to learn; most people will live in one. Still, for terminal-native developers it is one of the more thoughtfully designed options.
What it can do
- Three modes: interactive TUI, one-shot CLI, and zsh-plugin (`:` prefix) operation
- 300+ models via provider configuration (Claude, GPT, Grok, DeepSeek, Gemini, ...)
- Custom agents, skills, and semantic workspace search
- Git integration and shell-command tools in zsh mode
- Conversation and session management
- Apache-2.0 licensed; Discord community
Setup
- Install:
curl -fsSL https://forgecode.dev/cli | sh(or grab a release binary). - Configure your provider credentials (Claude, OpenAI, Gemini, etc. — see the provider configuration docs).
- Launch interactive mode, use one-shot CLI flags, or install the zsh plugin and prefix prompts with
:. - Tune agents, skills, and workspace search in the advanced configuration.
Prerequisites
A Unix-like shell (Linux/macOS); API credentials for at least one supported model provider.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (metrics tier, metrics-v1). Evidence: Adoption 7: starScore from 7,638 stars on tailcallhq/forgecode (log scale vs 298k peer max); Maintenance 9: pushed 2026-10-15 (recency 10); very active development; Openness 10: Apache-2.0 (0.7*10) + official docs verified (0.3*10); Manual 7: real terminal agent with distinctive zsh-plugin mode, corroborated by README; docked: '300+ models' is provider-config breadth (provider-claimed), smaller team. License: Apache-2.0.
★ 7,638 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Adoption 7: starScore from 7,638 stars on tailcallhq/forgecode (log scale vs 298k peer max)
- · Maintenance 9: pushed 2026-10-15 (recency 10); very active development
- · Openness 10: Apache-2.0 (0.7*10) + official docs verified (0.3*10)
- · Manual 7: real terminal agent with distinctive zsh-plugin mode, corroborated by README; docked: '300+ models' is provider-config breadth (provider-claimed), smaller team
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.