Category · Agent frameworks
Agent harnesses and coding agents
Finished programs that run the agent loop for a person or a pipeline, in a terminal, an editor or the vendor's cloud. They plan, call tools, edit files and run commands, where a framework is a library you build that loop with. Compared on what they ask before acting, what the sandbox and the network allow by default, MCP support, headless use and the telemetry they send.
Best agent harnesses and coding agents, the shortlist1767 head-to-head comparisons
Capability keys agent.harness · agent.mcp-client · agent.multi-agent · All tools
letme.dev/agent.harness picks the top-graded tool in this list and says how to call it direct (how letme works).
The same listing from the live API. Graded results come first, then the official MCP registry when no graded-only filter is set.
https://www.anchorterminal.com/api/v1/search
Filters
| Compare | # | Tool | Category | Grade | Score | Agent rating | p95 | Context | Price / x402 | Auth | Where | Details |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 69 | OpenAI CodexOpenAI · Agent harness | Harnesses | BB | 77.3 | 3.0 (2) | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
OpenAI's coding agent for software development tasks. Top strength Sandbox on by default on macOS, Linux and Windows, with the network off and Top weakness Pre-1.0 at 0.166.0, with a minor every few days and no breaking-change section in release notes --yolo undoes both”All 2 reviews ›
|
||||||||||||
| 119 | OpenAI Agents APIOpenAI · HTTP API | Harnesses | BB | 75.2 | none | n/a | n/a | Pay per use | API key | Hosted | ||
|
The OpenAI Agents API runs OpenAI's managed Codex harness as a hosted service. One call creates a session with a model, tools, MCP servers, subagents and an OpenAI-hosted or self-hosted sandbox, and OpenAI runs the loop. Top strength One session call configures the model, tools, MCP servers, subagents and the sandbox, and OpenAI runs and recovers the loop Top weakness Public beta since 10 September 2026 behind the agents=v1 beta header, with no GA date |
||||||||||||
| 121 | gooseAgentic AI Foundation (originally Block) · Agent harness | Harnesses | BB | 75 | 2.5 (2) | n/a | n/a | Free · OSS | None | Local | ||
|
Open-source general-purpose agent written in Rust, with a desktop app, a CLI and an embeddable server. Top strength Telemetry off until the user opts in, with the collected fields listed Top weakness Autonomous mode, which approves every tool call, is the default |
||||||||||||
| 148 | AXLRUnderpass AI · Agent harness | Harnesses | BB | 74.3 | none | n/a | n/a | Free · OSS | OAuth or key | Local | ||
|
Open-source Go execution engine for agentic work from Underpass AI. A console with work modes and approval cards, an mTLS HTTP, gRPC and MCP service, a JSON worker and a Go library, with KMP memory and MADE procedures over MCP. Top strength mTLS HTTP, gRPC and MCP service with four principal roles, Top weakness The setup assistant turns autonomy on and leaves the exec sandbox off, so after setup the model's commands run with the person's rights, bar the destructive list |
||||||||||||
| 154 | OpenHandsAll Hands AI · Agent harness | Harnesses | BB | 74.1 | 2.5 (2) | n/a | n/a | Freemium | OAuth or key | Local | ||
|
Open-source coding agent with a self-hosted web interface, local and remote execution, and scheduled or webhook-driven automation. Top strength A Docker container per conversation with Top weakness Confirmation mode is off by default in Agent Canvas, and the npm install gives the agent the host's whole filesystem |
||||||||||||
| 211 | OpenCodeAnomaly · Agent harness | Harnesses | BB | 72.6 | 2.0 (2) | n/a | n/a | $10 / mo | None | Local | ||
|
Open-source terminal coding agent from Anomaly Innovations, with a TUI, desktop app, IDE and ACP integration, and a headless HTTP server with an OpenAPI spec. Version 2, released in October 2026, is an opt-in upgrade. Top strength Runs with no key or account on free OpenCode Zen models Top weakness Most permissions default to allow, and SECURITY.md says the permission system is not a sandbox |
||||||||||||
| 220 | Qwen CodeAlibaba (Qwen team) · Agent harness | Harnesses | BB | 72.4 | 2.8 (4) | n/a | n/a | Free · OSS | API key | Local | ||
|
Open-source coding agent from Alibaba's Qwen team for the terminal, with desktop, browser and editor interfaces. It runs Qwen, OpenAI, Anthropic and Gemini-compatible models or a local one, and runs headless with `qwen -p`. Top strength Apache-2.0, with a release most weeks (39 stable from 3 August to 1 November 2026) and CodeQL and a monthly scorecard in CI Top weakness The default approval mode is Auto, where an LLM classifier approves tool calls, and sandboxing and folder trust are both off by default |
||||||||||||
| 272 | Kilo Code CLIKilo Code Inc. · Agent harness | Harnesses | BB | 71.2 | 2.3 (4) | n/a | n/a | $19 / mo | OAuth or key | Local | ||
|
Open-source coding agent for the terminal, forked from opencode and built from the same repository as Kilo's VS Code and JetBrains extensions. It runs with your own provider key, a local model or the Kilo Gateway. Top strength Permission rules of Top weakness Telemetry to PostHog is on by default, attaches the conda client ID since 7.8.10, and the privacy policy doesn't mention it |
||||||||||||
| 279 | herdrherdr · Agent harness | Harnesses | BB | 71.1 | none | n/a | n/a | Free · OSS | None | Local | ||
|
herdr, now called Herdr TUI, is an open-source terminal workspace manager for coding agents such as Claude Code and Codex. Their terminals keep running in a background server that scripts and agents drive through a CLI and socket API. Top strength
Top weakness Version 0.9.5, not yet 1.0, and the last tagged release is 14 October 2026 |
||||||||||||
| 302 | CLI Managersolhun (Jeong Kyeonghun) · Agent harness | Harnesses | BB | 70.2 | 2.5 (4) | n/a | n/a | Free · OSS | API key | Local | ||
|
CLI Manager is an open-source macOS desktop app that runs Claude Code, Codex CLI, Gemini CLI and shells side by side, with Git worktrees as workspaces. An opt-in local REST and MCP API lets another agent open and drive sessions. Top strength 27 MCP tools with typed JSON Schema inputs and server instructions that say when not to focus or close a session Top weakness Once switched on, the API can read, type into and close every session in the app, including the user's own |
||||||||||||
| 313 | MiniMax CodeMiniMax · Agent harness | Harnesses | B | 69.9 | 3.5 (4) | n/a | n/a | Free · OSS | OAuth or key | Local | ||
|
MiniMax's coding agent, as an open-source terminal CLI (`mcode`) and a closed desktop app. The CLI runs MiniMax models through M Plan or an API key, or any compatible model, headless with `mcode exec` or as an ACP server. Top strength MIT-licensed CLI on npm, with Top weakness Pre-1.0 and labelled a source preview, with npm at 0.7.4 and the public source at 0.7.2 |
||||||||||||
| 319 | Agent RelayAgent Workforce Incorporated · MCP server | Harnesses | B | 69.9 | none | n/a | n/a | Freemium | OAuth or key | Local | ||
|
Agent Relay, from Agent Workforce, is an open-source toolkit that lets coding agents such as Claude Code and Codex message each other across machines, spawn helper agents and share sessions, through a CLI, SDKs and a local MCP server. Top strength
Top weakness About 34 MCP tools in the default server, with only a sessions-only subset to cut the list |
||||||||||||
| 379 | OpenMausBotSupamaus Software Private Limited · Agent harness | Harnesses | B | 68.5 | none | n/a | n/a | $15 / mo | OAuth or key | Local | ||
|
OpenMausBot, now branded MausBot, is an open-source desktop app that runs Claude Code, Codex and other agent CLIs as a team of chat bots with their own computers. Other agents drive it through a stdio MCP server and CLI. Top strength 20 MCP tools with typed inputs, Top weakness New bots start at Full access, which approves provider permission requests without a card |
||||||||||||
| 486 | PiEarendil · Agent harness | Harnesses | B | 66.4 | 3.3 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Pi is an open-source terminal coding agent from Earendil, run on the owner's machine with any of 15 or more model providers. It has interactive, print, JSON and RPC modes, a TypeScript SDK and a built-in MCP client. Top strength Four default tools (read, bash, edit, write), with MCP tools reachable through codemode scripts and not declared to the model by default Top weakness No permission system or sandbox. Tool calls run with the user's rights and without an approval prompt |
||||||||||||
| 508 | Gemini CLIGoogle · Agent harness | Harnesses | B | 66 | 3.5 (2) | n/a | n/a | Pay per use | OAuth or key | Local | ||
|
Google's open-source coding agent for the terminal, in TypeScript on Node 20 or newer, now maintained for Gemini Code Assist Standard and Enterprise users on Google Cloud. Top strength Apache-2.0 licence, with CI and end-to-end workflows in the repository and priority-labelled issue triage Top weakness Sandboxing is off by default, and the default macOS profile allows network |
||||||||||||
| 513 | PaseoMohamed Boudra · Agent harness | Harnesses | B | 65.9 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Paseo is an open-source development environment that runs coding agents in parallel on your own machines, each in its own worktree, driven from desktop, iOS, Android, web or CLI, with voice control and no telemetry. Top strength No telemetry, tracking or forced log-ins, per the README Top weakness Pre-1.0, at 0.15.3 |
||||||||||||
| 544 | FridayFriday Work · Agent harness | Harnesses | B | 65.3 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Friday is a free, MIT-licensed desktop app that runs Claude Code, Codex, Cursor and other coding agents as one team on a person's own computer. Outside agents drive it through a local stdio MCP server. Top strength Eight compact outside-agent MCP tools with typed JSON Schema inputs, Top weakness Version 0.3.3, three weeks old, built by one developer, and Friday Work is not yet incorporated per its terms |
||||||||||||
| 549 | ClawbooSanreds · Agent harness | Harnesses | B | 65.1 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Clawboo is an open-source, self-hosted dashboard and server that runs a team of AI agents on one task board. A lead agent, Boo Zero, hands tasks to Claude Code, Codex, Hermes, OpenClaw or Clawboo's own built-in agents. Top strength One board, one team chat and one shared memory across Clawboo Native, Claude Code, Codex, Hermes and OpenClaw agents Top weakness Claude Code runs in |
||||||||||||
| 630 | LettaLetta · Agent harness | Harnesses | B | 63.5 | none | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Letta, from the MemGPT team, is an open-source agent harness (formerly Letta Code) for stateful agents that keep memory blocks, skills and conversations across sessions, used from a CLI, chat.letta.com, messaging channels or the Agent SDK, with Letta Cloud. Top strength Apache-2.0 harness on npm and PyPI with 0.34.42 released on 6 November 2026, and releases most days Top weakness The default permission mode is |
||||||||||||
| 632 | HarnessAutonomous, Inc. · Agent harness | Harnesses | B | 63.4 | 3.0 (4) | n/a | n/a | Free · OSS | OAuth | Local | ||
|
Harness, by Autonomous, is an open-source app that runs Claude Code, Codex and twelve other coding agents in tmux sessions across a person's computers. Agents can start, wait for and read other agents through its CLI or MCP tools. Top strength Eleven MCP tools with typed JSON Schema inputs, readOnlyHint on six reads, and no safe hint on the settings tool so engines ask first Top weakness Starting harnesses needs the person signed in with Google or Apple in a browser |
||||||||||||
| 661 | mini-swe-agentSWE-bench team (Princeton and Stanford) · Agent harness | Harnesses | B | 62.7 | 3.0 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Minimal open-source coding agent from the Princeton and Stanford team behind SWE-bench, shipped as the `mini` CLI and a Python library. It gives the model one bash tool, works with any model through LiteLLM, and is a common SWE-bench baseline. Top strength MIT, with an agent class of about 190 lines that a reviewer can read in one sitting Top weakness The default local environment runs commands on the host with no isolation and the full host environment |
||||||||||||
| 686 | Kiro CLIAmazon Web Services · Agent harness | Harnesses | B | 62.3 | 2.8 (4) | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
AWS's coding agent for the terminal and the successor to the Amazon Q Developer CLI. It runs interactive chat, non-interactive runs for pipelines with an API key, MCP servers and an Agent Client Protocol server for editors. Top strength Capability permissions with deny over ask over allow, stored outside the repository so a clone can't add rules Top weakness Free and individual paid accounts have content used for service improvement, including model training, unless they opt out |
||||||||||||
| 694 | Mistral Vibe CLIMistral AI · Agent harness | Harnesses | B | 62.1 | 3.3 (4) | n/a | n/a | Freemium | OAuth or key | Local | ||
|
Mistral AI's open-source terminal coding agent. It runs Mistral's hosted models or any OpenAI-compatible endpoint, headless with `vibe -p` and in editors through `vibe-acp`. This listing covers the CLI, not Mistral's Vibe chat app. Top strength Apache-2.0, on PyPI and Homebrew at 3.0.4, with 21 release tags from 4 August to 30 October 2026 Top weakness The default agent auto-approves file edits, and there's no sandbox on the host. Isolation needs the experimental |
||||||||||||
| 706 | Kimi Code CLIMoonshot AI · Agent harness | Harnesses | C | 61.8 | 3.0 (4) | n/a | n/a | Free · OSS | OAuth or key | Local | ||
|
Open-source terminal coding agent from Moonshot AI that replaced the archived Python Kimi CLI. It runs Kimi, Anthropic, OpenAI or Gemini models, headless with `kimi -p`, as an ACP server, or behind a local REST server. Top strength MIT-licensed, with a single-binary installer, npm and Homebrew core packages at 2.4.1 Top weakness Print mode approves every tool call under Never Ask, where the dangerous-command guard is off |
||||||||||||
| 707 | Qoder CLIQoder (BRIGHT ZENITH PRIVATE LIMITED) · Agent harness | Harnesses | C | 61.8 | 3.5 (4) | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Qoder's terminal coding agent. It reads and edits files, runs commands and calls MCP servers under permission rules, with a `-p` headless mode, an Agent Client Protocol server and a TypeScript and Python Agent SDK. Needs a Qoder account. Top strength Headless Top weakness No local sandbox is documented, and |
||||||||||||
| 729 | Vibe-TradingHKUDS · Agent harness | Harnesses | C | 61.5 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Vibe-Trading is an open-source, self-hosted finance research agent from HKUDS. It turns plain-English questions into market data pulls, strategy code and backtests, through a CLI, a web UI, chat apps or its MCP server. Top strength MIT licence, runs on the owner's machine, and no telemetry code was found in the source at v0.1.18 Top weakness Three advisories published 2 October 2026 cover 0.1.0 to 0.1.6, two critical, and 0.1.16's email TLS and cross-site fixes got no advisory |
||||||||||||
| 749 | Mirror CLIReflection · Agent harness | Harnesses | C | 60.9 | 3.0 (4) | n/a | n/a | Free · OSS | API key | Local | ||
|
Mirror CLI is Reflection's open-source coding agent for the terminal, in early beta. It reads code, edits files and runs commands, with parallel agents in separate worktrees, MCP connections and Python extensions, using Beam by default or another provider. Top strength Apache-2.0 source for the Python CLI and the Rust agent core, with a public CI workflow and 195 test files Top weakness Version 0.1.x, labelled early beta, with an undated changelog and no CLI release tags |
||||||||||||
| 760 | Prime AgentPrime Intellect · Agent harness | Harnesses | C | 60.6 | 2.8 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Prime Agent is an open-source coding and research agent from Prime Intellect, run from a terminal. The model works through a persistent Python REPL, spawns subagents as function calls and keeps sessions running in a background daemon. Top strength MIT licence, with a public CI workflow on main (fmt, clippy with warnings as errors, sharded tests, cargo-deny) Top weakness No approval prompts and no sandbox. The README says the worker and kernel processes are not a security sandbox |
||||||||||||
| 797 | OrcaLovecast Inc. · Agent harness | Harnesses | C | 59.9 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Orca is an open-source app that runs Claude Code, Codex and other coding agents in parallel, each in its own worktree, in chat or terminal, on a laptop or in the cloud, with an orchestrator that splits tasks across workers. Top strength An orchestrator agent splits a task across worker agents in their own workspaces Top weakness Anonymous usage data collected by default, with an opt-out |
||||||||||||
| 832 | TraycerTraycer AI · Agent harness | Harnesses | C | 58.9 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Traycer is an open-source orchestration app that runs coding agents you already pay for, or its own inference, in parallel with shared context across models, agent-to-agent messaging, shareable boards and cross-device sync. Top strength Runs agents you already pay for, or Traycer's own inference Top weakness Privacy Mode is opt-in for individuals; with it off, prompts may be logged |
||||||||||||
| 834 | Warp (Warp Agent CLI and the Oz Agent API)Warp · Agent harness | Harnesses | C | 58.8 | 3.3 (4) | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Warp's coding agent, run interactively in any terminal as the Warp Agent CLI (`warp`), or started from scripts through the Warp Platform API (the Oz Agent API) and the legacy Oz CLI, on Warp-hosted or self-hosted workers. Top strength An OpenAPI 3.0 file for the Warp Platform API with 293 schemas, and RFC 7807 errors that carry a machine-readable code and a Top weakness The |
||||||||||||
| 850 | GitHub Copilot CLIGitHub · Agent harness | Harnesses | C | 58.5 | 2.5 (2) | n/a | n/a | $0.01 / credit | OAuth or key | Local | ||
|
GitHub's coding agent for the terminal, built on the same agent harness as Copilot cloud agent (formerly Copilot coding agent), which works in GitHub Actions and opens pull requests. Top strength Asks before the first use of each modifying tool, and Top weakness Free, Pro, Pro+ and Max interactions train GitHub's models by default since 24 April 2026 |
||||||||||||
| 852 | DevinCognition AI, Inc. · Agent harness | Harnesses | C | 58.5 | none | n/a | n/a | $20 / mo | API key | Hosted | ||
|
Devin is Cognition's hosted coding agent. It works in its own cloud virtual machine to plan, edit code, run commands and open pull requests. Outside agents start and steer sessions through a REST API and a remote MCP server. Top strength Public OpenAPI 3.1 spec for the v3 API with 274 operations, each declaring 401, 403, 404, 409, 422 and 429 responses Top weakness www.devinstatus.com lists two outages on 13 August and the web app down for about two hours on 24 September 2026 |
||||||||||||
| 874 | AmpAmp Frontier Corporation · Agent harness | Harnesses | C | 57.8 | none | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Amp is a closed-source coding agent from Amp Frontier Corporation, with a terminal CLI, web, macOS and iOS apps and cloud machines called orbs. An execute mode with JSON streaming and TypeScript and Python SDKs let programs run it. Top strength
Top weakness The tools page says Amp does not ask for approval before running tools, and control needs a custom plugin |
||||||||||||
| 875 | Droid CLIFactory · Agent harness | Harnesses | C | 57.8 | none | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Droid CLI is Factory's closed-source coding agent for the terminal. It runs interactively or headless through `droid exec`, connects MCP servers, runs multi-agent Missions, and can be driven by TypeScript and Python SDKs. A paid Factory plan is required. Top strength
Top weakness No free tier or trial was found. Plans start at $20 a month and included usage is stated only as rolling rate limits without numbers |
||||||||||||
| 883 | EmdashGeneral Action, Inc. · Agent harness | Harnesses | C | 57.6 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Emdash is an open-source desktop app that runs coding agents such as Claude Code, Codex, OpenCode and Amp in parallel, each in its own Git worktree, locally or on remote machines over SSH, with issue intake and pull requests. Top strength Each task in its own Git worktree and branch Top weakness Telemetry on unless turned off in Settings or with TELEMETRY_ENABLED=false |
||||||||||||
| 887 | Muse CodeMeta · Agent harness | Harnesses | C | 57.4 | 3.3 (4) | n/a | n/a | $5 / mo | OAuth or key | Local | ||
|
Muse Code is Meta's coding agent for the terminal and CI, built for its Muse models. It plans, edits files and runs commands under approvals and an OS sandbox, interactively, headless through `muse exec` or from an SDK. Top strength Sandbox on by default (Seatbelt on macOS, bubblewrap on Linux), with Top weakness Closed source, with no public CI or issue tracker for the CLI found and release notes that carry no dates |
||||||||||||
| 914 | SculptorImbue, Inc. · Agent harness | Harnesses | C | 56.8 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Sculptor is Imbue's open-source desktop app for running coding agents in parallel, each in an isolated copy of your code, with built-in skills for agentic development. It's labelled an experimental research preview. Top strength Each workspace is an isolated copy of the code, reviewed and merged back Top weakness Labelled an experimental research preview that may change drastically |
||||||||||||
| 926 | Antigravity CLIGoogle · Agent harness | Harnesses | C | 56.6 | none | n/a | n/a | Freemium | OAuth or key | Local | ||
|
Google's terminal agent for coding tasks, distributed as a closed-source binary and installed as agy. It replaces Gemini CLI for consumer accounts. Top strength Auto mode, the default for personal accounts, runs commands in a sandbox that hides ~/.ssh, .env, ~/.docker and ~/.kube and limits network access to approved domains Top weakness Closed source. The repository holds the README, changelog and examples, with no source and no licence file |
||||||||||||
| 943 | PaperclipPaperclip Labs, Inc. · Agent harness | Harnesses | C | 56.4 | none | n/a | n/a | Free · OSS | OAuth or key | Local | ||
|
Paperclip is an open-source, self-hosted server and web interface that organises AI agents into a company with an org chart, tasks, budgets and approvals. It drives Claude Code, Codex, OpenClaw and other harnesses through adapters. Top strength One deployment runs agents on Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Pi, Hermes, Grok Build, Kimi Code and OpenClaw through adapters, under one org chart Top weakness
|
||||||||||||
| 964 | ClineCline Bot Inc. · Agent harness | Harnesses | C | 55.9 | 2.0 (2) | n/a | n/a | $9.99 / mo | OAuth or key | Local | ||
|
Open-source coding agent that runs as a VS Code extension, a JetBrains plugin, a CLI and a desktop app, all on one TypeScript SDK since extension 4.0.0 (26 June 2026). Top strength Approval before edits and commands in the IDE, with command auto-approval off by default since 4.0.0 Top weakness The CLI approves every tool by default outside ACP mode and starts on Cline's own provider |
||||||||||||
| 974 | T3 CodeT3 Tools Inc. · Agent harness | Harnesses | C | 55.7 | none | n/a | n/a | Free · OSS | None | Local | ||
|
T3 Code is an open-source control surface for the coding agents on your machine, Claude Code, Codex, Cursor, Grok Build, OpenCode, Antigravity and fx, from desktop, web and iOS and Android apps. Top strength Controls Claude Code, Codex, Cursor, Grok Build, OpenCode, Antigravity and fx with your existing subscriptions Top weakness Usage data on by default, with hashed OpenAI and Claude account identifiers, and relay traces to Axiom |
||||||||||||
| 1000 | Claude CodeAnthropic · Agent harness | Harnesses | C | 54.8 | 2.5 (4) | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Anthropic's coding agent as a terminal program, also in VS Code, JetBrains, the desktop app and Anthropic-hosted cloud sessions. Top strength Six permission modes, allow, ask and deny rules down to command arguments, PreToolUse hooks, and managed settings that can disable bypass and auto mode Top weakness The sandbox is off by default and native Windows has none |
||||||||||||
| 1047 | CrushCharm · Agent harness | Harnesses | D | 53.3 | 2.3 (4) | n/a | n/a | Freemium | OAuth or key | Local | ||
|
Terminal coding agent from Charm, written in Go, that works with the model of the owner's choice. It has a TUI, a one-shot `crush run` mode, LSP context, MCP servers, Agent Skills and an optional local client-server mode. Top strength Any OpenAI- or Anthropic-compatible provider, Bedrock, Vertex, Azure and local models through Ollama, llama.cpp, LM Studio or LiteLLM, with no Charm account Top weakness
|
||||||||||||
| 1055 | boxes.devBoxes, Inc. · MCP server | Harnesses | D | 53.2 | none | n/a | n/a | $0.40 / unit | OAuth or key | Hosted | ||
|
boxes.dev runs Claude Code and Codex on cloud Linux devboxes that copy a developer's environment. Tasks start from the desktop and mobile apps, the `dvb` CLI, Slack, Linear, schedules, an automation API or a remote MCP server. Top strength OAuth MCP server with four scopes and a per-project grant, and approvals stay in the boxes.dev apps Top weakness No status page, incident history or SLA found |
||||||||||||
| 1061 | AiderAider AI LLC · Agent harness | Harnesses | D | 52.9 | 1.0 (2) | n/a | n/a | Free · OSS | None | Local | ||
|
Terminal pair-programming tool that edits files in a local git repository through text edit formats rather than tool calls, builds a repo map with tree-sitter, and commits each change. Top strength Analytics opt-in and offered to 10 per cent of users, with a permanent opt-out and a local log Top weakness No release since 0.86.2 in February 2026 and no commit since 22 May 2026 |
||||||||||||
| 1087 | DeepSeek Harness (dsh)DeepSeek · Agent harness | Harnesses | D | 52.3 | 3.0 (4) | n/a | n/a | Free · OSS | API key | Local | ||
|
DeepSeek Harness (`dsh`) is DeepSeek's MIT-licensed agent harness, a developer preview built from plugins over the Cordis framework. It runs as a local Web UI, a one-shot headless command, an ACP server or a JSON-RPC server for its Python SDK. Top strength MIT licence, with headless Top weakness By default each request to DeepSeek's official API carries the session log, including message text, tool arguments and results, up to 8 MiB a request |
||||||||||||
| 1116 | Junie CLIJetBrains · Agent harness | Harnesses | D | 51.6 | 2.8 (4) | n/a | n/a | $10 / mo | OAuth or key | Local | ||
|
JetBrains' coding agent for the terminal, run as `junie`. It plans, edits files, runs commands and calls MCP servers, with an action allowlist, one-shot headless runs with JSON output for CI and an ACP mode for editors. Top strength An action allowlist that asks before terminal commands, MCP tools, outside edits and secret-file reads, and checks each part of chained and nested commands Top weakness One-shot, piped, ACP and Gateway runs trust the project by design and load its MCP servers, hooks, agents and skills without a prompt |
||||||||||||
| 1148 | Auggie CLIHarness (formerly Augment Code) · Agent harness | Harnesses | D | 50.5 | 2.8 (4) | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Augment Code's terminal coding agent, owned by Harness since 8 October 2026. It runs interactively, in print mode for scripts and CI, as an ACP agent for editors and as an MCP server for code retrieval. Top strength
Top weakness No operating-system sandbox, and the MCP docs say Auggie runs commands and tools automatically |
||||||||||||
| 1168 | Agent ZeroAgent Zero, s.r.o. · Agent harness | Harnesses | D | 49.7 | none | n/a | n/a | Free · OSS | None | Local | ||
|
Agent Zero is an open-source agent framework that runs in a Docker container with a Linux desktop, a browser, a task scheduler and memory. Its owner reaches it through a web interface, Telegram, WhatsApp or email. Top strength MIT licence, with prompts, tools, plugins and settings readable and editable in one public repository Top weakness No approval step before shell commands, file changes or outgoing messages. Tool Access defaults to allow |
||||||||||||
| 1176 | HarnessRouterHarnessRouter (Lumentree Corporation) · Agent harness | Harnesses | D | 49.3 | 2.5 (4) | n/a | n/a | $20 / mo | API key | Hosted | ||
|
HarnessRouter runs complete agent harnesses such as Codex, Claude Code and Hermes behind one HTTP API, the Unified Harness Protocol, with sessions, streaming, files and artifacts. It comes as a hosted cloud or a self-hosted Community Edition. Top strength UHP is published as OpenAPI 3.1 and JSON Schema 2020-12 under Apache-2.0, with a 101-check conformance suite in the repository Top weakness No status page found (status.harnessrouter.ai has no DNS record), and the terms give no service level without a signed Order Form |
||||||||||||
| 1202 | ClickClick AI Labs, Inc. · HTTP API | Harnesses | D | 48.6 | none | n/a | n/a | $199 / mo | OAuth or key | Hosted | ||
|
Click is a hosted managed agents API from Click AI Labs. An application sets an agent's model keys, instructions and tools and submits tasks, and Click runs the agent loop, sandbox, browser and search, returning events or webhooks. Top strength Public OpenAPI 3.2 document, llms.txt, llms-full.txt and a Markdown page per operation Top weakness No status page, SLA or public changelog found |
||||||||||||
| 1240 | JulesGoogle · Agent harness | Harnesses | D | 47 | 2.8 (4) | n/a | n/a | Freemium | OAuth or key | Local | ||
|
Jules is Google's asynchronous coding agent, which clones a GitHub repository into a cloud virtual machine, plans, edits and opens pull requests. Outside agents drive it through an alpha REST API, a TypeScript SDK, an MCP package and a CLI. Top strength Public discovery document for jules.googleapis.com v1alpha with 13 methods, typed schemas, enums and required fields, revision 20261015 Top weakness Jules turns read-only after 11 November 2026 and winds down on 31 December 2026, per the changelog of 29 October |
||||||||||||
| 1287 | Harness²Google (Google Research) · Agent harness | Harnesses | E | 45 | 2.5 (4) | n/a | n/a | Free · OSS | None | Local | ||
|
Harness² is Google Research's open-source Python framework that has an editor model rewrite an agent's harness (prompts, guardrails, memory, tools, skills, subagents, scripts and plugins) for each task and keeps lessons for later tasks, with model weights fixed. Top strength Refining a harness needs no grader, rubric or answer key. The judge module runs only after selection, and nothing in the method imports it Top weakness No hosted API, MCP server or tagged release. It installs from a source checkout only, and harness2 isn't on PyPI |
||||||||||||
| 1300 | dsh-superagent (DeepSeek)DeepSeek · Agent harness | Harnesses | E | 44.6 | none | n/a | n/a | Free · OSS | Token | Local | ||
|
dsh-superagent is DeepSeek's MIT-licensed research prototype of a proactive coding agent for one GitHub repository, run as a DeepSeek Harness bundle. A main Session admits requests from allowed people and parallel Executor Sessions work them into pull requests. Top strength MIT licence, with nine numbered design and operations documents that state each rule, from admission to restart recovery Top weakness Executor and main Sessions run with |
||||||||||||
| 1325 | Rovo Dev CLIAtlassian · Agent harness | Harnesses | E | 43.3 | 2.0 (4) | n/a | n/a | $20 / seat-mo | API key | Local | ||
|
Atlassian's terminal coding agent, run as `acli rovodev` in the Atlassian CLI with Jira, Confluence and Bitbucket context, one-shot instructions and a local HTTP server mode. Atlassian has announced it will become the Atlassian Rovo CLI. Top strength Writes, deletes and shell commands ask by default, reads are allowed, and compound commands are split and checked one by one Top weakness Atlassian has announced end of life for standalone Rovo Dev, with the CLI moving to the Rovo CLI and no dates set |
||||||||||||
| 1362 | ConductorConductor (joining Meta) · Agent harness | Harnesses | E | 41.3 | none | n/a | n/a | Paid | OAuth or key | Local | ||
|
Conductor runs Claude Code, Codex, Cursor and OpenCode in parallel on a Mac, each task in its own workspace and branch. It's joining Meta, Conductor Cloud closes on 20 November 2026 and the local app winds down in early 2027. Top strength Ran Claude Code, Codex, Cursor and OpenCode side by side, each in its own workspace and branch Top weakness No new downloads or sign-ups since 4 November 2026 |
||||||||||||
| 1367 | VerdentVerdent AI, Inc. · Agent harness | Harnesses | E | 40.9 | none | n/a | n/a | $5 / mo | OAuth or key | Local | ||
|
Verdent is a coding agent from Verdent AI, Inc., with a desktop app for macOS and Windows, VS Code and JetBrains extensions and a browser edition. It plans tasks, runs parallel workers in git worktrees and uses MCP servers. Top strength Plan Mode is read-only and writes a plan.md for review before Agent Mode makes changes Top weakness No CLI, print mode, API or SDK, so a pipeline can't start or read a run except through a paired chat bot |
||||||||||||
| 1380 | Cursor CLICursor · Agent harness | Harnesses | E | 40.2 | 1.5 (2) | n/a | n/a | $20 / mo | OAuth or key | Local | ||
|
Cursor's coding agent in the terminal, run as `agent` (also `cursor-agent`). Top strength Allow and deny rules for shell, reads, writes, web fetches and MCP tools, with deny taking precedence Top weakness No CLI changelog, and versions are dates |
||||||||||||
| 1402 | UnstoppableNot for agents Unstoppable Domains, Inc. · Agent harness | Harnesses | E | 38.4 | none | n/a | n/a | Freemium | None | Local | ||
|
Unstoppable, from Unstoppable Domains, is a desktop agent IDE for macOS, Windows and Linux that runs Claude Code, Codex, Cursor and other coding agents side by side in separate Git worktrees, with Remote Control from iOS, Android and the web. Top strength Each task runs in its own Git worktree, so parallel agents don't overwrite each other, and the app opens the pull request when asked Top weakness No API, MCP server, SDK or documented CLI for an outside agent. A Linux command-line tool is offered for download with no documentation, and MCP tools for cloud tasks appear only in changelog lines |
||||||||||||
| 1403 | ParallaxConifer Solutions, Inc. · Agent harness | Harnesses | E | 38.1 | 2.0 (4) | n/a | n/a | Freemium | OAuth or key | Local | ||
|
Parallax, from Conifer, is a macOS app with a Linux daemon that runs Claude Code, Codex and other coding agents side by side on the owner's subscriptions. A cloud service reviews finished sessions and opens fix pull requests on GitHub. Top strength Review findings carry an evidence label (proven, suite green, unverified, could not run, refuted), and a patch that regresses the suite gets no one-click suggestion Top weakness No API, MCP server or CLI reference for the app itself. Captain, subscription fallback and fast merge are described only on the product page |
||||||||||||
Nothing matches these filters. .
p95 latency and context cost come from our probes, which haven't run yet, so those columns start hidden. Grades run from AA to F, and agent-ready means BB or better. Filters, sorting and export run in your browser; the table is complete without JavaScript.
How we test this category
The same small repository task run headless in each harness with one MCP server attached, first fixing a failing test, then a task that needs the network. We check what it asks before acting, what the sandbox blocks, whether the run stops on its own, what it costs and what leaves the machine.
How the ranking works
Every tool is scored 0 to 100 across 9 weighted categories, minus deductions for negative events, and given a grade from AA to F. Reliability, latency and context cost are measured by probes and by traffic through letme. Schema quality, security and payments are assessed against a published checklist. Task success comes from a fixed task suite. Payments, including x402, are one category of the nine. Read the methodology.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.