CopilotKit
by CopilotKit (incorporated as Tawkit Inc.)
The frontend stack that puts your AI agents in front of real users
CopilotKit is an MIT-licensed frontend stack for agentic applications. It takes an agent you already built in LangGraph, CrewAI, Mastra, PydanticAI or the Claude Agent SDK and gives it a production chat interface, generative UI, shared application state and human-in-the-loop approvals across React, Angular, Vue, Slack, Teams and mobile - the plumbing most agent teams otherwise rebuild by hand for every surface.
CopilotKit is an open-source SDK that occupies the layer almost every agent project discovers it needs and nobody wants to build: the horizontal frontend between an agent and its users. The premise is that agent frameworks solved orchestration but left teams hand-rolling streaming chat UIs, tool-call rendering, approval prompts and state synchronisation for each surface separately. CopilotKit supplies those as React, Angular, Vue and React Native components plus a lightweight runtime you mount on your own backend as a Next.js, Express, NestJS, Bun, Deno or Cloudflare Workers endpoint, so agent traffic stays server-side. It is deliberately backend-agnostic, with documented adapters for LangGraph (Python, FastAPI and TypeScript), CrewAI Flows, Mastra, PydanticAI, AWS Strands, Google ADK, Microsoft Agent Framework, LlamaIndex, AG2, Agno and the Claude Agent SDK, alongside its own built-in agent. Its four load-bearing capabilities are generative UI, where an agent renders real application components rather than markdown; shared state, keeping agent and UI in sync in both directions; frontend tool execution; and human-in-the-loop approval gates with an auditable action record. The company also authors AG-UI, an open protocol standardising agent-to-user-interface communication that Google, Microsoft, AWS and Oracle have adopted - the strategic bet being that the interface layer becomes a standard rather than a vendor. Founded by brothers Atai and Uli Barkai and originally shipped as a Show HN in July 2023, the project has 37.3k GitHub stars under an MIT licence, roughly 25 staff, and raised $27M total ($6.5M seed plus a $20.5M Series A announced 5 May 2026) from Glilot Capital Partners, NFX, SignalFire and 97212 Ventures. Commercially it sells Copilot Cloud, a hosted layer for threads, retention, channels and analytics; the editor-side SDK itself stays free.
The product engineering team that has a working agent and now has to ship it as a real feature inside an existing React or Angular application, across web, Slack and mobile, without writing the UI layer three times.
Agentic chat, generative UI, shared state and human-in-the-loop approvals as installable components, so one agent reaches every surface without a bespoke frontend per channel.
At a Glance
- Category
- Agent Development
- Pricing
- Freemium, Subscription, Contact for pricing
- Target Market
- CTOs, Enterprise Developers, Frontend Engineers, AI Product Managers, Startup Founders
- Deployment
- Open-source, Cloud-first, Self-hosted, Hybrid
- Founded
- 2023
- Headquarters
- Seattle, United States
- Team Size
- 11-50
Key Features
- ✓Generative UI
Agents render real interactive application components on demand rather than returning markdown, so the agent's output is part of your product's interface.
- ✓Shared state
Keeps agent state and application UI synchronised in both directions, so the agent can read and mutate what the user is currently looking at.
- ✓Human-in-the-loop approvals
Explicit approval gates and frontend tool execution with an auditable action record before an agent performs a consequential operation.
- ✓Multi-surface channels
One agent deploys to React, Angular, Vue, React Native, Slack and Microsoft Teams, with Discord, WhatsApp, Telegram and SMS listed as coming.
- ✓AG-UI protocol
An open standard for agent-to-user-interface communication, authored by CopilotKit and adopted by Google, Microsoft, AWS and Oracle.
- ✓Framework-agnostic runtime
A lightweight server endpoint mounted on Next.js, Express, NestJS, Bun, Deno or Cloudflare Workers that keeps all agent communication server-side.
- ✓Broad backend adapters
Documented support for LangGraph, CrewAI Flows, Mastra, PydanticAI, AWS Strands, Google ADK, Microsoft Agent Framework, LlamaIndex, AG2 and Agno.
Capabilities
Use Cases
- •In-app copilot for a SaaS product
A B2B application adds an assistant that reads the user's current view, drafts changes and asks for confirmation before applying them.
- •Agent-driven dashboards
An analytics agent generates chart and table components on the fly rather than describing results in prose the user then has to reconstruct.
- •Approval-gated automation
A finance workflow lets an agent prepare a payment run but blocks execution until a named human approves it, leaving an auditable record.
- •One agent across web and Slack
A support agent serves the web app and Slack from a single implementation, so behaviour and context stay identical across both channels.
- •Migrating an existing LangGraph prototype
A team with a working notebook-stage LangGraph agent ships it as a product feature without rewriting orchestration or hand-building streaming UI.
Ideal For
Best For
- ✓Embedding an existing LangGraph, CrewAI or Claude Agent SDK agent into a React or Angular product as a first-class in-app feature
- ✓Generative UI, where the agent renders real interactive application components instead of returning a wall of markdown
- ✓Human-in-the-loop workflows that need an explicit approval gate and an auditable record before an agent takes a consequential action
- ✓Shipping one agent across web, Slack and Microsoft Teams without maintaining a separate integration and UI per channel
- ✓Teams that want to avoid interface lock-in by building against the open AG-UI protocol rather than a single vendor's chat widget
Not Ideal For
- ✗Teams that only need a streaming chat box - the Vercel AI SDK or assistant-ui are lighter, and CopilotKit's shared-state and generative-UI machinery is overhead you will not use
- ✗Python-only shops with no JavaScript frontend: the runtime is a JS backend proxy, and Chainlit builds a conversational UI directly from Python instead
- ✗Organisations that require a SOC 2 report or ISO 27001 certificate before adopting a vendor - no public attestation was found, and there are no G2, Capterra or TrustRadius listings to corroborate enterprise claims
- ✗Teams expecting every feature in the MIT repo to be free - Hacker News commenters report that parts of the open-source codebase require an enterprise licence
Integrations
Deployment
Market Analysis
Pros
- ✓MIT-licensed with 37.3k GitHub stars, so the core is genuinely adoptable without a contract or a sales conversation
- ✓Unusually broad framework support - LangGraph, CrewAI, Mastra, PydanticAI, AWS Strands, Google ADK, Microsoft Agent Framework, Claude Agent SDK, LlamaIndex, AG2 and Agno
- ✓AG-UI adoption by Google, Microsoft, AWS and Oracle is meaningful third-party validation of the interface layer it is standardising
- ✓Cheap to evaluate: a free forever developer tier and a $39/month Pro tier before any seat-based commitment
- ✓Runtime mounts on your own backend, so agent traffic and credentials stay server-side rather than transiting a vendor proxy by default
Cons
- ✗Hacker News commenters report that parts of the open-source codebase require an enterprise licence, citing basic 'on error' handlers - the MIT label does not mean every feature is free
- ✗Self-hosting is heavier than it looks: the runtime is a JS backend proxy holding chat state, which one practitioner called 'quite a pain if you want to scalably self-host'
- ✗The 'Copilot' name is badly overloaded against GitHub Copilot and Microsoft Copilot, and HN readers said they could not work out who was behind the project
- ✗No public SOC 2, ISO 27001 or HIPAA attestation was found, and there is no G2, Capterra or TrustRadius listing to sanity-check the enterprise logos on the site
- ✗Roughly 25 staff against Fortune 500 logos - support depth and roadmap durability are a real diligence question at that ratio
Pricing
Developer
$0
- ✓Free forever, one developer
- ✓200 max threads, 3-day retention, 1 GB storage, 500 channel credits
Pro
From $39/mo
- ✓5,000 max threads, 5-day retention, 10 GB storage
- ✓2,000 monthly channel credits
Team
From $100/seat/mo
- ✓Up to 5 seats, 25,000 max threads, 14-day retention, 100 GB storage
- ✓Optional VPC / on-prem deployment with self-hosting database included
Enterprise
Contact for pricing
- ✓Unlimited threads, custom retention, VPC / on-prem included
- ✓Dedicated engineering support up to 5 hrs/week, SLA, priority bug fixes, early-access analytics and self-learning
The SDK is MIT-licensed and free; what is metered is Copilot Cloud, priced on conversation threads, history retention, storage and monthly channel credits rather than on tokens or messages. Developer is free forever for a single developer at 200 threads and 3-day retention; Pro is $39/month; Team is $100 per seat per month for up to five seats and is the first tier where VPC or on-prem deployment is included; Enterprise is quote-only and adds unlimited threads, an SLA and up to five hours a week of dedicated engineering support. Note that Hacker News commenters report some features inside the open-source repo still require an enterprise licence, so 'MIT' should not be read as 'every capability is free'.
Security & Compliance
Connect
Sources
This page was written from 7 sources, 5 on domains other than copilotkit.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Mastra
Open-source TypeScript framework and platform for building, deploying and observing production AI agents and workflows
Agentrys
Agentic design automation that lets chip teams build and own a self-improving engineering workforce
OpenAI Agents API
The Codex agent harness behind one API call — managed sessions, sandboxes and subagents
Letta
Open-source platform for stateful AI agents with persistent memory, now with a TypeScript Agents SDK for embedding them in your own apps