Automation · v1.1.2 · MIT
Local CLI & Task Automation
Creates idempotent build scripts, git hooks, and reproducible local task runners with dry-run support.
Decision & Local Setup Blueprint Engine · Privacy-First
Tell AgentPilot what you want to accomplish. Discover suitable AI models, relevant Agent Skills, compatible local clients, workflow steps, and estimated token costs before you start.
No account required · Local evaluation · Zero hidden telemetry

Conceptual pipeline: every recommendation inspects real catalog provenance, client compatibility, and single-agent sufficiency before generating a key-free local setup package.
01
User Task
Objective + Budget
02
Planner
Local Engine
03
Model
187 Cataloged
04
Skills
32 Audited
05
Workflow
Acyclic DAG
06
Est. Cost
Min / Exp / Max
The Problem
Teams waste time and budget bouncing between models, unnecessary multi-agent loops, and unverified community skills without knowing cost or compatibility upfront.
01.
Frontier, economy, reasoning, and local open-weight models vary widely in context limits, tool calling, and per-token pricing.
02.
Claude Code, OpenAI Codex CLI, Cursor, GitHub Copilot, Gemini CLI, and OpenCode expect different paths and configuration files.
03.
Splitting a straightforward task across multiple agents adds context handoff tokens, coordination steps, and extra retries.
04.
Outdated prices and unscanned SKILL.md repositories expose local workspaces to budget surprises and supply-chain risks.
How AgentPilot Works
AgentPilot is a decision and local installation platform—not a cloud agent host or API proxy.
State what you need to build, analyze, or automate alongside optional budget and client constraints.
The local planner checks required capabilities: coding, reasoning, vision, long context, or offline privacy.
Matches suitable models, minimal viable agent count, and compatible Open Agent Skills (SKILL.md).
Displays transparent Min, Expected, and Max ranges broken down by planner, execution, handoff, and retry tokens.
Inspect destination paths, automatic backups, permissions, SHA-256 hashes, and hardware checks before writing anything.
The Python Local Runtime installs configs and verified skills on your machine with one-click disk verification and rollback.
Model Intelligence
Compare real models across cloud APIs and local open-weight runtimes. Stale or unknown pricing is always flagged—never invented.
| Model | Provider | Context Window | Input / 1M | Output / 1M | Pricing Source | Spec |
|---|---|---|---|---|---|---|
| AX-Qwen3.5-0.8B-decision-MLX-AXQ-MXFP4-MTP huggingface/automatosx-ax-qwen3.5-0.8b-decision-mlx-axq-mxfp4-mtp | Qwen · local | 131k | $0 | $0 | community reported · 2026-11-21 | View Model Spec |
| AX-Qwen3.5-0.8B-decision-MLX-AXQ-MXFP8-MTP huggingface/automatosx-ax-qwen3.5-0.8b-decision-mlx-axq-mxfp8-mtp | Qwen · local | 131k | $0 | $0 | community reported · 2026-11-21 | View Model Spec |
| AX-Qwen3.5-9B-MLX-AXQ-6bit-MTP huggingface/automatosx-ax-qwen3.5-9b-mlx-axq-6bit-mtp | Qwen · local | 131k | $0 | $0 | community reported · 2026-11-21 | View Model Spec |
| AX-Qwen3.5-9B-MLX-AXQ-MXFP4-MTP huggingface/automatosx-ax-qwen3.5-9b-mlx-axq-mxfp4-mtp | Qwen · local | 131k | $0 | $0 | community reported · 2026-11-21 | View Model Spec |
| AX-Qwen3.5-9B-MLX-AXQ-MXFP8-MTP huggingface/automatosx-ax-qwen3.5-9b-mlx-axq-mxfp8-mtp | Qwen · local | 131k | $0 | $0 | community reported · 2026-11-21 | View Model Spec |
| AX-Qwen3.6-27B-MLX-AXQ-6bit-MTP huggingface/automatosx-ax-qwen3.6-27b-mlx-axq-6bit-mtp | Qwen · local | 131k | $0 | $0 | community reported · 2026-11-21 | View Model Spec |
Open Agent Skills
Every skill exposes its source provenance, SHA-256 checksum, required permissions, compatible clients, and static security status.
Automation · v1.1.2 · MIT
Creates idempotent build scripts, git hooks, and reproducible local task runners with dry-run support.
Backend Development · vgit-49f948f · unknown
Build, review, refactor, or architect ASP.NET Core web applications using current official guidance for .NET web development. Use when working on Blazor Web Apps, Razor Pages, MVC, Minimal APIs, controller-based Web APIs, SignalR, gRPC, middleware, dependency
Backend Development · v1.3.2 · MIT
Designs idempotent REST and RPC endpoints, input schema validation, rate limiting, and structured error envelopes.
Backend Development · vgit-e4bb5d9 · Complete terms in LICENSE.txt
Reference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the
Backend Development · vgit-49f948f · unknown
Build a composable CLI for Codex from API docs, an OpenAPI spec, existing curl examples, an SDK, a web app, an admin tool, or a local script. Use when the user wants Codex to create a command-line tool that can run from any repo, expose composable read/write c
Backend Development · vgit-49f948f · unknown
Generates custom design system rules for the user's codebase. Use when user says "create design system rules", "generate rules for my project", "set up design rules", "customize design system guidelines", or wants to establish project-specific conventions for

Cost & Topology Discipline
Adding extra agents serializes context across handoffs and increases token usage. AgentPilot recommends a single agent by default and explains the exact trade-offs when role separation is justified.
Shared context window, zero inter-agent handoff tokens, simpler debugging for focused implementation or research.
Reserved for workflows where independent security auditing, compliance review, or multi-domain verification justifies extra tokens.
Estimates are transparent planning ranges based on catalog token pricing and workload assumptions—not guaranteed final provider bills.
Run Single vs Multi-Agent ChallengeLocal-First Installation
Your prompts, provider credentials, and private files stay strictly local. AgentPilot never asks for API keys on the web and never proxies inference.
Important: Cloud model APIs are configured as local client references only—cloud model weights are never downloaded locally. Local open-weight models list explicit RAM/disk requirements and manual pull commands (e.g., Ollama).

Open the interactive planner directly—no account registration, no API key form, and no tracking scripts.