Beyond a simple terminal chat loop. Npxstack Forge orchestrates the entire engineering lifecycle: headless task runs (forge run), repository preflight inspection, automated test & self-healing regression repair, persistent SQLite ledgers, financial budget controls, and hybrid offline Ollama + Cloudflare Edge models.
irm https://cli.npxstack.dev/forge/install.ps1 | iexforge in any repository root.npm: @npxstack/forge@0.2.2Switch scenarios to see autonomous batch runs, test self-healing, stack inspection, and interactive pair-programming.
Forge is not just an interactive terminal chat. It is an end-to-end autonomous engineering platform covering headless tasks, stack inspection, test-driven repair, and financial cost governance.
Execute end-to-end coding tasks headlessly in scripts or CI/CD pipelines without manual prompts. Forge autonomously investigates the repository, designs a structured plan, edits files, and delivers verified results.
Never commit broken code. Forge automatically discovers your test runner (Vitest, Jest, PyTest, Go test), executes the test suite, parses failure stack traces, and initiates iterative repair cycles until all tests pass.
Deep automated preflight of your codebase. Automatically detects frameworks (Next.js, Node, Python, Go, Rust), package managers, scripts, and protected assets, while diagnosing system readiness.
For engineers who love the shell. An interactive terminal session with multi-turn context memory, in-terminal syntax-highlighted git diffs, and instant slash commands to control every aspect of your session.
Absolute protection for proprietary code. Run 100% offline via local Ollama with zero network egress, or tap into high-throughput cloud models (Qwen 2.5 Coder 32B, DeepSeek R1, Llama 4 Scout 131k context) across 300+ edge cities.
Enterprise observability and cost control. Every run, prompt, model call duration, and git checkpoint is recorded in a local SQLite ledger, with strict financial spending caps to prevent runaway cloud bills.
Engineered for high autonomy, verifiable test execution, local-first safety, and zero cloud lock-in.
Autonomous headless execution
Multi-turn pair-programming
100% offline, zero cloud egress
Qwen 2.5 Coder, DeepSeek R1, Llama 4 Scout
Inference prompts are processed strictly in ephemeral memory and discarded immediately upon response streaming. We do not train models on your proprietary code.
Set hard spending limits per task run with --max-cost and daily/monthly caps to guarantee full financial control and avoid unexpected cloud bills.
Every single autonomous run, prompt, model decision, token consumption metric, and git checkpoint is recorded in an auditable local SQLite database on your machine.
Manage your active authentication token, inspect live edge gateway connectivity, and monitor quota.
Generate a new token for this workstation, or import an existing token if you already generated one on another device.
Create an instant cryptographic HMAC token for your device to authenticate Forge CLI with the Cloud Gateway.
Run fully offline locally with Ollama for $0, or tap into high-throughput cloud models via Cloudflare Workers AI.
Ideal for solo builders, hobbyists, and offline local development.
For engineers building real-world projects needing flagship reasoning and coding models.
For development teams needing shared quotas, pooled ledgers, and priority support.
Switch models dynamically in your terminal using /model <alias> or arrow keys.
| Model Name | Category | Short Alias | Context Window | Function Calling |
|---|---|---|---|---|
| Qwen 2.5 Coder 32BFlagship Coder | Code | coder | 32,768 tokens | Native Tools |
| DeepSeek R1 Distill Qwen 32BReasoning | Code / Reasoning | deepseek, r1 | 32,768 tokens | Native Tools |
| Llama 4 Scout 17B MoE131k Context | Agent | scout, llama4 | 131,000 tokens | Native Tools |
| Llama 3.3 70B Fast Agent | Agent | llama3, llama | 32,768 tokens | Native Tools |
| Kimi K2.7 Code (Moonshot) | Code | kimi | 32,768 tokens | Native Tools |
| OpenAI GPT-OSS 120B | Agent | gpt | 65,536 tokens | Native Tools |
| Qwen 2.5 Coder 7B (Local)100% Offline | Local (Ollama) | local, qwen | Local Context | Local Tools |
Everything you need to know about Forge terminal agent, offline security, and cloud edge models.