OpenAI AgentKit alternatives (2026)

Free to self-host and cheapest at scale; billed per full run, not per credit.

A flat $20/mo platform fee with unlimited runs (AI tokens metered separately) makes budgeting more predictable than OpenAI AgentKit's per-credit burn.

Built around multi-agent "teams" with a transparent dual meter.

Biggest free tier (5,000 credits/mo) and strong for data-heavy flows.

The most hands-off "AI employee" with voice/phone, if you want done-for-you.

Botpress is a no-code AI agent and chatbot builder pairing a visual Studio with a code-first developer kit (ADK). It deploys agents to channels like WhatsApp, Messenger, Slack and web chat, and bundles the underlying LLM cost into a per-conversation price with no markup. Pricing is per conversation, billed in monthly packs once you exceed your plan allowance.

Voiceflow is a no-code/low-code platform for designing, building, and deploying AI chat and voice agents on a visual canvas. It targets agencies and businesses that want to ship customer-facing assistants across web, voice, and messaging channels without writing code, while keeping control over LLM choice, knowledge bases, and tools. Pricing moved to a unified credit model (April 2025) where a monthly plan fee includes a credit allowance and usage is metered against it.

Stack AI is a no-code, drag-and-drop platform for building enterprise AI agents and workflows that connect LLMs to internal data sources, documents, and apps. It leans hard on governance: on-prem/VPC deployment, SSO, access control, and SOC 2, HIPAA, and GDPR compliance. It targets regulated teams (finance, healthcare, legal) that need agents without writing code.

Dify is an open-source platform for building LLM apps, agentic workflows and RAG pipelines through a visual canvas, with a hosted cloud tier and a free self-hostable Community Edition. It blends a no-code/low-code builder with backend-as-a-service APIs, native MCP (client and server), and integration with hundreds of model providers, so teams can ship chatbots, agents and workflows without wiring infrastructure from scratch.

Taskade is an all-in-one AI productivity workspace (tasks, docs, mind maps, chat) that doubles as a no-code builder for custom AI agents and automations. You assemble agents, deployable apps, and multi-step automations against your project data, and the platform bills on a tiered plan that bundles a fixed monthly allotment of "AI credits" consumed by agent and automation runs across 15+ models (GPT, Claude, Gemini).

Text.com is the parent company behind LiveChat (agent-driven live chat helpdesk) and ChatBot (a no-code visual bot builder), now converged into one per-seat support suite. ChatBot lets non-developers build automated chat flows with a drag-and-drop builder plus an AI Agent trained on your content, while LiveChat handles human agent handoff, visitor tracking, and proactive campaigns. Both bill per agent seat and bundle a small monthly allotment of AI resolutions, charging per-resolution overages beyond it.

v0 is Vercel's AI-powered, prompt-to-app builder that turns natural-language prompts (and design images) into working React/Next.js + Tailwind UIs and full-stack apps. It blends a no-code chat canvas with one-click Vercel deployment and GitHub sync, but bills on token-metered credits rather than seats alone, so it sits between a no-code builder and a real coding tool.

Bubble is a full-stack visual no-code platform for building production web and mobile apps without writing code, covering UI, database, and backend workflows. It moved to a workload-based pricing model: plans bundle a monthly allowance of "workload units" (WU) that meter every backend operation, and usage beyond the allowance is billed as overage. It is a builder for finished apps, not an AI agent platform; AI capabilities exist via plugins and its API connector.

CustomGPT.ai is a no-code platform for turning your own content (1,400+ file types plus connectors like Google Drive, SharePoint, Notion, Zendesk and HubSpot) into custom RAG chatbots/agents you can embed, share, or call via API. It targets businesses that want an accurate, hallucination-resistant support or knowledge agent without engineering, and bills on a per-query model rather than per-seat.

Softr is a no-code platform for building client portals, internal tools, and web apps on top of a database (its own Softr Database, Airtable, Google Sheets, or external REST APIs). It targets non-developers who assemble apps from pre-built blocks with permissions, user groups, and payments, rather than writing code or wiring a backend. Increasingly AI-forward: it ships an AI app builder and a hosted MCP server that lets AI tools read/write Softr data.

Glide is a no-code platform that turns spreadsheets and databases (Glide Tables, Google Sheets, Airtable, Excel, SQL) into custom mobile and web business apps. It has pivoted hard toward AI: the Glide Agent can generate a working app from a prompt, and you can embed AI actions (extract data, draft emails, transform text/audio/images) into app workflows. Pricing is a flat monthly plan plus usage-based "updates," where most data writes to external sources and API/workflow actions are metered.

As of mid-2026 vellum.ai hosts "Vellum: Your Personal Intelligence" — an open-source (MIT) personal AI assistant marketed as an "AI chief of staff for high-agency individuals." It lives in Vellum Cloud (or self-hosted), has its own identity/email, a persistent multi-type memory layer, and acts in the world: triages email, manages calendar, posts to Slack/X, updates Linear, makes real phone calls, runs web search, and even vibe-codes small web apps. Capabilities extend via installable "skills" (a catalog at skills.sh incl. Gmail, Slack, DoorDash, Amazon) and via user-added MCP servers (Linear, GitHub, Notion, etc.). NOTE: this is NOT the older Vellum LLMOps/prompt-engineering-and-workflows developer platform that previously occupied this domain; that product appears to have been replaced/repositioned, so any "Pro $500/mo, prompt/workflow execution limits" pricing you find for "Vellum" is stale.