This is a sourced directory of the tools you drive models with. Each row names the maker and links to the site or public repo when one exists. The local-endpoints marker shows which harnesses can target hardware you own, which is how a harness connects to the towers side of this planner.
Rows were last checked on 2026-08-05. Capabilities change; treat this as a starting map and confirm endpoint support against each project’s current docs before you wire anything up.
Free and open source; you pay whichever model provider you point it at.
Self-improving TUI harness launched August 5, 2026. A persistent IPython REPL is the only tool, sub-agents are function calls, and the agent can edit its own prompts, skills, and memory at runtime. Works with open and closed frontier models; a background daemon keeps sessions attachable and recoverable. It runs model-generated Python with your user permissions and ships no sandbox, so give it the same trust you would give a script you have not read.
Claude Pro/Max subscription or Anthropic API keys.
Terminal agent with first-party Anthropic models. Custom base URL / gateway setups are documented in the wild; confirm against current Anthropic docs before relying on a self-hosted endpoint.
Cursor subscription tiers; some flows also accept provider API keys.
Agentic IDE. Custom model settings exist, but IDE traffic may leave the machine, so localhost OpenAI-compatible endpoints are not counted as supported here. Cursor CLI support for custom base URLs is separate and partial.
Free control surface; you pay for whichever provider CLIs you already run.
GUI and remote control plane over installed providers such as Claude Code, Codex, Cursor CLI, and OpenCode. It does not talk to models directly; local-endpoint support depends on the backend provider you attach.
GitHub Copilot Individual, Business, or Enterprise subscription.
Separate from VS Code itself. Agent and chat features run through GitHub’s hosted model stack; self-hosted OpenAI-compatible endpoints are not a supported path for the product UI.
Bring hosted, subscription-backed, gateway, or local models.
Local-first personal agent gateway across chat apps. Broader than a pure coding CLI, but commonly used to drive coding work and custom model endpoints. Public source; confirm the current LICENSE file before redistributing.
Google Antigravity product access; SDK is open for building on the harness.
Agentic IDE, CLI, and SDK from Google. Product agent traffic uses Google’s stack. The published SDK is open; the IDE itself is not a self-hosted OpenAI-compatible client.
Editor is free/open; Zed AI and third-party models bill separately.
Open editor with an agent panel that can use hosted Zed AI or custom OpenAI-compatible providers. Most source is GPL-3.0; some reusable crates are Apache-2.0.
Warp Free, Pro, and Business plans cover Warp Agent usage.
Agentic terminal product with Warp Agent built in. Terminal source has been opened; agent model routing remains product-hosted. Confirm current docs before assuming a self-hosted OpenAI-compatible path.
Open agent; you pay whichever LLM provider you configure.
Local-first agent from Block with CLI and desktop surfaces plus MCP extensions. Stewardship now sits with the Agentic AI Foundation; confirm the current canonical org on GitHub before linking forks.
Self-host free; OpenHands Cloud is paid. Inference is BYOK or cloud-bundled.
Open software-agent platform with Agent Canvas UI, SDK, and self-hosted or cloud backends. Supports OpenAI-compatible providers when you configure them. Confirm the current LICENSE file before redistributing.
Google account free tier or Gemini / Vertex API billing.
Open terminal agent for Gemini models with tools, MCP, and headless modes. Defaults to Google hosted APIs; treat arbitrary local OpenAI-compatible bases as unverified.
Xcode is free; ChatGPT and Claude integrations use those accounts. Custom and locally hosted chat providers are configurable in Intelligence settings.
Agent and chat surfaces built into Xcode. Apple documents internet-hosted providers and locally hosted Chat Completions endpoints under Intelligence settings.
Open self-host path is BYOK; Plandex Cloud is optional.
Terminal agent oriented around multi-file plans and accumulated diffs. Configure providers including OpenAI-compatible bases. Confirm the current LICENSE file before redistributing.