AI Agent Workflow
Pick the right agent runtime for your workflow. Compare licenses, billing, model support, MCP, subagents, and per-million-token pricing across the five tools -- verified against vendor docs for May 2026.
Pro/Max sub covers daily use. Best subagent + skill UX.
MIT, 372k★, runs across 20+ chat platforms. The lobster way.
MIT, 153k★. Self-improving skill loop + 7 deploy backends.
Cron-triggered skills with SSE web UI + AGENT_OS_CAPS gating.
OpenClaw fork wired to Claude OAuth. Free messaging assistant -- ToS gray zone.
Feature Matrix
14 rows. All claims sourced from vendor docs as of 2026-05-16. Hover GitHub stars verified via GitHub API on the same date.
| Attribute | Claude Code | OpenClaw | Hermes Agent | Custom OpenClaw (myseodesk fork) | agent-os |
|---|---|---|---|---|---|
| Vendor | Anthropic | openclaw (OSS) | Nous Research | Greg / myseodesk | Greg (myseodesk) |
| License | Proprietary | MIT | MIT | MIT (fork of openclaw) | Internal |
| Self-host | No | Yes | Yes | Yes | Yes |
| Billing | Pro $20/mo · Max 5x $100/mo · Max 20x $200/mo + API | free (BYO model API) | free (BYO model API) | free (rides Claude Code OAuth) | free (uses Claude OAuth) |
| Models supported | Anthropic only (Opus 4.7, Sonnet 4.6, Haiku 4.5) | Provider-agnostic (OpenAI, Anthropic, OpenRouter, local) | 200+ via OpenRouter, OpenAI, Anthropic, HF, NVIDIA NIM, custom | Claude via OAuth token + any BYO provider | Inherits Claude Code (Anthropic) via OAuth token |
| MCP support | Yes | Yes | Yes | Yes | Yes |
| Subagents | Yes | Yes | Yes | Yes | Yes |
| Skills / memory | Yes | No | Yes | No | Yes |
| Scheduled jobs | Yes | Yes | Yes | Yes | Yes |
| Messaging platforms | -- | WhatsApp, Telegram, Slack, Discord, iMessage, +15 more | Telegram, Discord, Slack, WhatsApp, Signal, CLI | WhatsApp, Telegram, Discord | -- |
| Run platforms | CLI, VSCode, JetBrains, Web | macOS menu bar, iOS, Android, Linux, Windows, Docker | Local, Docker, SSH, Modal, Daytona, Vercel Sandbox, Singularity | Self-hosted Node 22+, Docker | Web dashboard, FastAPI, systemd cron |
| GitHub ★ | -- | 372,355 | 152,958 | -- | -- |
| Ideal for | Daily driver on Anthropic stack | Local-first personal assistant across messaging apps | Self-improving agent with persistent skill memory across platforms | Personal multi-channel assistant on existing Claude sub | Headless scheduled Claude Code skill runs with report aggregation |
| Weakness | Locked to Anthropic models. Sub doesn't cover 3rd-party SDK apps. | Younger ecosystem, setup heavier than hosted CLIs. | Heavier infra story. Python-first, less polished IDE integration. | Rides Claude Code OAuth -- Anthropic ToS gray zone. Token revoke risk if enforcement tightens. | Single-user, tied to operator's Claude OAuth + skill library. |
Workflow Visualization
Each tool routes the same user request through a different stack. Mapping below shows where you'd plug in models and where the agent runs.
Decision Wizard
Three questions. Answer honestly about constraints, get one tool back.
Pro $20/mo, Max $100, Max 20x $200. API option for SDK use. Polished UX, native MCP + subagents + skills.
MIT, free self-host. 20+ messaging platforms, voice, live canvas. BYO model API. TypeScript / Node 22+.
MIT, free self-host. Self-improving skill loop, 7 deploy backends (local / Docker / Modal / Vercel Sandbox / etc). 200+ models via OpenRouter.
Fork of openclaw wired to CLAUDE_CODE_OAUTH_TOKEN.
Routes WhatsApp + Telegram + Discord traffic to your Pro/Max sub. Free at point of use.
Caveat: Anthropic ToS gray zone -- third-party SDK use of subscription quota is unsupported.
Token revoke or account ban possible if enforcement tightens.
Internal tool. Cron-triggered Claude Code skills with SSE web UI, report aggregation, and AGENT_OS_CAPS gating. Inherits Claude OAuth.
Cost Calculator
Adjust monthly token volume and in/out split. Compare per-model API costs. Ollama Pro shown as flat-sub baseline.
Sub-as-API
Filtered list. Each plan below lets you sign in once and route the sub's quota into an agent CLI (Claude Code, Codex, Gemini CLI, Copilot CLI, or any agent that hits the local Ollama endpoint). App-only subs without an agent-CLI sign-in path (ChatGPT Go, Cursor Pro, Perplexity Pro, Mistral Le Chat Pro, xAI Premium+) are excluded.
| Plan | Vendor | Monthly | Annual | Agent CLI (sub login) | Works in third-party agent? | Notes | Verified |
|---|---|---|---|---|---|---|---|
| Claude Pro | Anthropic | $20/mo | $17/mo | Claude Code (claude setup-token) | Gray zone -- CLAUDE_CODE_OAUTH_TOKEN works in other agents, ToS unclear | Annual billing $204/yr = $17/mo. Token piping into non-Claude-Code agents = ToS gray. | 2026-05-16 |
| Claude Max 5x | Anthropic | $100/mo | -- | Claude Code (5x Pro quota) | Gray zone | Best $/agent-hour for heavy Claude Code users. | 2026-05-16 |
| Claude Max 20x | Anthropic | $200/mo | -- | Claude Code (20x Pro quota) | Gray zone | Top tier. Effective per-hour rate best at 8+ hr/day Claude Code use. | 2026-05-16 |
| ChatGPT Plus | OpenAI | $20/mo | -- | Codex CLI (limited Plus quota) | No -- Codex token scoped to Codex | Codex CLI sign-in works with Plus, but Plus quota throttles agent runs. | UNVERIFIED -- openai.com blocked WebFetch; cross-referenced aionx 2026-02-25 |
| ChatGPT Pro | OpenAI | $200/mo | -- | Codex CLI (full quota) | No -- Codex token scoped to Codex | OpenAI's Claude-Code analogue. Sub pipes into Codex CLI via sign-in. | UNVERIFIED -- openai.com blocked; cross-referenced aionx 2026-02-25 |
| Google AI Pro | $19.99/mo | -- | Gemini CLI (Google sign-in mode) | Limited -- Gemini CLI only | Rebranded from Gemini Advanced. Gemini CLI sub-login mode wires sub quota into CLI; bundles 2TB storage + Workspace AI. | UNVERIFIED -- gemini.google.com blocked; cross-referenced aionx 2026-02-25 | |
| Google AI Ultra | $249.99/mo | -- | Gemini CLI (Google sign-in, highest quota) | Limited -- Gemini CLI only | Top consumer tier launched 2026. Includes Veo/Imagen highest tiers + 30TB. | UNVERIFIED -- gemini.google.com blocked; cross-referenced aionx 2026-02-25 | |
| GitHub Copilot Pro | GitHub | $10/mo | -- | Copilot CLI + IDE (GitHub login) | No -- Copilot token scoped to Copilot | Cheapest agent-CLI sub. GitHub switching Copilot to usage-based metering 2026-06-01. | 2026-05-16 |
| GitHub Copilot Pro+ | GitHub | $39/mo | -- | Copilot CLI + IDE (full quota) | No -- Copilot token scoped to Copilot | Includes Claude/Gemini model picker inside Copilot. | 2026-05-16 |
| Ollama Pro | Ollama | $20/mo | $16.67/mo | Any agent that hits local Ollama API | Yes -- OpenAI-compatible local endpoint, unrestricted | Only sub where the API is the product. 12 cloud models incl. DeepSeek V4 Pro 1.6T. Annual $200/yr = $16.67/mo. | 2026-05-16 |
| Ollama Max | Ollama | $100/mo | -- | Any agent that hits local Ollama API | Yes -- highest cloud-model quota | Heavy cloud-model users. Same OpenAI-compatible local endpoint. | 2026-05-16 |
Caveat: "ToS gray zone" entries (Claude OAuth token in non-Claude-Code agents like Custom OpenClaw) work in practice today but are unsupported per Anthropic ToS. Token revoke or account ban possible if enforcement tightens.
Sources: vendor pricing pages (verified 2026-05-16 where possible); cross-referenced with aionx.co/ai-comparisons/ai-pricing-comparison/ (updated 2026-02-25). Vendor pages that block crawlers (OpenAI, Google, Perplexity) marked UNVERIFIED -- re-check before purchase.
Model Pricing Reference
Frontier-tier API models and one subscription baseline. Every row links to the vendor's source page so you can re-verify on the day you actually buy.
| Model | Vendor | Input $/M | Output $/M | Context | Best for | Source |
|---|---|---|---|---|---|---|
| Claude Opus 4.7 | Anthropic | $5 | $25 | 1M | Hardest reasoning + agentic coding | link |
| Claude Sonnet 4.6 | Anthropic | $3 | $15 | 1M | General coding + workflow execution | link |
| Claude Haiku 4.5 | Anthropic | $1 | $5 | 1M | Cheap bulk classification + drafts | link |
| GPT-5.5 | OpenAI | $5 | $30 | 272K | General frontier (price doubles above 272K ctx) | link |
| GPT-5.5 Pro | OpenAI | $10 | $45 | 272K | Hardest OpenAI reasoning | link |
| DeepSeek V4 Pro | DeepSeek | $1.74 $0.435 promo | $3.48 $0.87 promo | 1M | Cheapest frontier reasoning (1.6T params) | link |
| DeepSeek V4 Flash | DeepSeek | $0.14 | $0.28 | 1M | Cheap fast reasoning at scale | link |
| Gemini 3.1 Pro | $3 | $15 | 2M | Deep research + long-context reasoning | link | |
| Gemini 3.1 Flash-Lite | $0.25 | $1.5 | 200K | Cheap content gen + summarization | link | |
| Kimi K2.6 | Moonshot | $0.6 | $2.5 | 256K | HTML/UI generation + vision | link |
| Qwen3-Coder 480B | Alibaba | $0.22 | $0.9 | 262K | Deep code review, second-opinion passes | link |
| GLM 5.1 | Zhipu | $1.05 | $3.5 | 203K | Agentic coding over long sessions | link |
| Ollama Pro (subscription) | Ollama | $20/mo | unlimited | 1M | Unlimited cloud access to 12 frontier models (DeepSeek V4 Pro, Qwen3-coder, Kimi, GLM, etc.) | link |