Forklane

Autonomous coding agent for your terminal.

Forklane is a terminal-based autonomous coding agent for long-running tasks and massive codebase development. It operates directly in your terminal with a rich TUI, connects to multiple LLM providers, and orchestrates multi-file changes through an intelligent pipeline that scales from one-line fixes to cross-cutting transformations spanning hundreds of files.

Install

npm install -g @forklaneai/forklane

Requires Node.js >= 22.19.0. No other dependencies — the CLI ships as a single self-contained bundle.

Usage

# Interactive mode (opens TUI)
forklane

# Alias
forklane-agent

# One-shot prompt for CI/CD
forklane -p "Add error handling to src/routes/users.ts"

# Interactive session
forklane
forklane> Refactor the authentication system to use OAuth 2.0

# Resume a previous session
forklane -S <session-id>

# Non-interactive with JSON output
forklane -p "Fix the type errors" --output-format json

First run launches an interactive onboarding wizard to pick your provider and model.

Core capabilities

• Specialist agent profiles — Editor (surgical code editing), Rewriter (full-file rewrites and refactoring), Searcher (codebase analysis), Auto (task-adaptive); switchable via /agent
• Four-tier memory — working memory with priority-tagged compaction, episodic recall (TF-IDF), long-term persistent memory, semantic vector recall (HNSW), plus a language-aware codebase graph for TS, JS, Python, Rust, Go, and Java
• Verification pipeline — 7 gates (syntax, lint, typecheck, unit tests, integration tests, architecture, full suite) with per-language auto-detection and self-correcting fix-and-reverify loops
• Multiple LLM providers — Anthropic, OpenAI, Google Gemini, Ollama (local, no API key), Kimi, and any OpenAI-compatible endpoint
• Forklane SSH remote execution — shift long-running workflows to a remote server; the daemon survives disconnects and your terminal streams events in real time
• Collaborative sessions — N users and N agents on one project through session hubs, with client-side visibility filtering and sealed credentials
• MCP client — stdio and HTTP/SSE transports
• Plugins and skills — manifest-based plugin discovery with lifecycle hooks; reusable prompt templates and inline scripts
• Session persistence — resumable JSONL sessions with crash recovery

CLI options

-S, --session <id>       Resume a session
-C, --continue           Continue last session
-y, --yolo               Auto-approve all actions
-m, --model <name>       Override LLM model
-p, --prompt <text>      Non-interactive mode
--plan                   Plan mode on startup
--auto                   Auto permission mode
--setup                  Re-run setup wizard
--output-format <fmt>    text | json | stream-json
--skills-dir <dir>       Additional skill directories
--cwd <path>             Working directory
--tui / --no-tui         Force/enable TUI

Configuration

Loaded in ascending priority: built-in defaults → ~/.forklane/config.toml → .forklane/config.toml (walked up from cwd) → environment variables.

FORKLANE_PROVIDER    LLM provider (anthropic, openai, ollama, etc.)
FORKLANE_MODEL       Model name
FORKLANE_API_KEY     API key for the provider
FORKLANE_BASE_URL    Custom base URL
FORKLANE_THINKING    Thinking level (none, low, medium, high)

Optional: local semantic embeddings

Semantic memory recall uses local embeddings via @xenova/transformers when it is available. It is not installed by default (it is a large dependency); Forklane degrades gracefully without it. To enable:

npm install -g @xenova/transformers

Links

• Website: forklane.ai

License

Proprietary. Copyright © 2026 Forklane (forklane.ai). All rights reserved.

This package is distributed for use with Forklane; it is not open source, and the source code is not licensed for redistribution or modification.