freshcrate
Skin:/
Home > Frameworks > intentkit

intentkit

IntentKit is an open-source, self-hosted cloud agent cluster that manages a collaborative team of AI agents for you.

Why this rank:Strong adoptionRelease freshnessHealthy release cadence

Description

IntentKit is an open-source, self-hosted cloud agent cluster that manages a collaborative team of AI agents for you.

README

IntentKit

IntentKit

Your Cloud-Native Agent Team

PyPI - License PyPI Version Docker Documentation

IntentKit is an open-source, self-hosted cloud agent cluster that manages a collaborative team of AI agents for you.

Why IntentKit

Current AI agents generally follow two paths:

  1. Local-First (e.g., OpenClaw): These serve as personal assistants or experimental tools. However, they often require expensive hardware and extensive local permissions.
  2. Cloud-Native: Similar to modern web apps, these run in the cloud, consume minimal local resources, require zero maintenance, and offer high reliability for completing tasks.

If you are looking for a robust, cloud-native solution, IntentKit is for you.

Features

  • â˜ī¸ Cloud-Native: Ultimate resource efficiency.
  • 🔒 Secure by Design: Agents are fundamentally unable to access any of your secret keys.
  • 🤖 Collaborative AI: Multiple agents that can call and interact with each other.
  • 🔄 Out-of-the-Box Ready: Fully configured and ready to use.
  • 🔗 Crypto-Friendly: Optional Web3 and blockchain integrations.
  • đŸĻ Social Media Integration: Seamlessly connects with social platforms.
  • đŸ› ī¸ Extensible Skill System: Easily add new capabilities.

Documentation

Please check out the Documentation before getting started.

Deployment

Read the Deployment Guide to get started with your setup.

Other Use Cases

Besides self-deploying IntentKit, you can also use the project in the following ways:

  • As a Python Library: Import and extend IntentKit, or use it to add agent cluster capabilities to your existing projects.
  • Via API: Regardless of the deployment method, you can interact with agents from external applications via the built-in API endpoints.

Contributing

Due to the rapid pace of AI development, we currently do not accept code contributions via Pull Requests. However, submitting feature requests and reporting bugs in the GitHub Issues section is highly appreciated and the best way to contribute.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Release History

VersionChangesUrgencyDate
v0.18.0# Release v0.18.0 ## New Features - WeChat agents now accept voice, video, and file messages in addition to images. Users can send an audio note, a short video, or a document and the agent will receive the content directly instead of a "type not supported" fallback. ## Improvements - Expanded the catalog of selectable LLM models — the latest DeepSeek, Claude Opus, Kimi, and MiMo releases are now available when configuring agents. - When an agent's underlying model returns an empty response (High4/24/2026
v0.17.60# Release v0.17.60 ## Bug Fixes - Fixed a long-standing issue where images sent through WeChat came through as unreadable garbage and the AI refused to process them. Inbound images now decrypt correctly, so users can share photos with their agents and get a real response. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.59...v0.17.60 High4/19/2026
v0.17.59# Release v0.17.59 ## Bug Fixes - Fixed an issue where images sent through WeChat could be stored incorrectly and cause the AI to reject them, with the error then blocking any further messages in the same conversation. Inbound images are now validated before being forwarded to the AI, and unrecognized or unsupported formats are dropped cleanly instead of poisoning the conversation. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.58...v0.17.59 High4/19/2026
v0.17.58# Release v0.17.58 ## Improvements - Improved reliability for agents with large skill catalogs. Built-in capabilities — current time, long-term memory, posts, activities, and sub-agent calls — now remain available even when the automatic tool-selection layer narrows the active tool set. Previously these core tools could be filtered out, causing agents to repeatedly attempt the same failed call without recovering. - Raised the threshold at which the automatic tool-selection layer activates and High4/19/2026
v0.17.57# Release v0.17.57 ## New Features - Added time-limited **share links** for chats and posts. Team members can now generate a public URL that lets anyone — including recipients without an account — view a chat transcript or post for three days. - When posts are delivered to WeChat or Telegram, the push message now contains a share link instead of a login-gated URL, so off-platform recipients can open the content directly. ## Improvements - Each public view increments a view counter on the shaHigh4/19/2026
v0.17.56# Release v0.17.56 ## Improvements - Tuned the threshold at which the automatic tool-selection layer activates, so agents with mid-sized skill sets now run without the extra selection step and only agents with larger skill catalogs trigger it. - Routine dependency updates. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.55...v0.17.56 High4/18/2026
v0.17.55# Release v0.17.55 ## Improvements - Added diagnostic logging that records which image URLs are forwarded to the model, making it easier to investigate image-processing errors in production. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.54...v0.17.55 High4/17/2026
v0.17.54# Release v0.17.54 ## Improvements - Silenced noisy Gemini-related warning logs (schema compatibility and automatic function calling notices) that were flooding production logs without adding operational value. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.53...v0.17.54 High4/17/2026
v0.17.53# Release v0.17.53 ## Bug Fixes - Fixed several issues in the WeChat integration that prevented inbound images from reaching the LLM and caused spurious "push failed" alerts even when messages were successfully delivered. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.52...v0.17.53 High4/17/2026
v0.17.52# Release v0.17.52 ## New Features - **Alert forwarding for Go integrations**: The Telegram and WeChat integration services now automatically forward error-level events to the configured alert channel (Telegram or Slack), matching the behavior already in place on the Python side. Both stacks share the same per-minute alert budget so operators are not flooded. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.51...v0.17.52 High4/17/2026
v0.17.51# Release v0.17.51 ## Bug Fixes - Fixed an issue where error-level events were not being forwarded to the configured alert channel (Telegram/Slack). Operators will once again receive automatic notifications when errors occur. ## Other - Updated dependencies. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.50...v0.17.51 High4/17/2026
v0.17.50# Release v0.17.50 ## New Features - **WeChat image storage**: Images received from WeChat are now downloaded, decrypted, and stored on S3, providing stable URLs that the AI model can reliably access. Previously, WeChat's temporary CDN links could expire before the model processed them. - **Cross-turn media forwarding**: The lead agent can now forward images, videos, and files to sub-agents even when the media was received in a previous message. Attachment URLs are made visible to the lead soHigh4/16/2026
v0.17.49# Release v0.17.49 ## New Features - **Media input support for LLM models**: Added audio, video, and file input support fields to model configuration, enabling richer media input options when selecting models. ## Bug Fixes - Fixed markdown rendering on post pages — headings and other typography styles were not displaying correctly. ## Other - Updated dependencies. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.48...v0.17.49 High4/16/2026
v0.17.48# Release v0.17.48 ## New Features - **WeChat image support**: Agents can now receive and process images sent by users through WeChat. Images are extracted from incoming messages and passed to the agent's model for understanding. ## Improvements - Improved error handling when users send images to agents using models that don't support image input — a clear message is now returned instead of a silent failure. - Simplified image capability detection for agents by removing the deprecated image High4/16/2026
v0.17.47# Release v0.17.47 ## Bug Fixes - Paid x402 actions now fail gracefully in agent conversations instead of surfacing as internal server errors when wallet funding or payment setup runs into trouble. ## Improvements - Improved Safe wallet x402 payment reliability with clearer timeout handling. - Added automatic administrator alerts when the x402 paymaster runs out of gas, so platform issues can be addressed faster. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.High4/15/2026
v0.17.46# Release v0.17.46 ## Improvements - Chat conversations now continue running on the server when you navigate to other pages — only the explicit Cancel button stops generation. Returning to the thread shows the latest messages, including ones produced while you were away. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.45...v0.17.46 High4/15/2026
v0.17.45# Release v0.17.45 ## Bug Fixes - Fixed a critical issue where autonomous (scheduled) tasks failed to run when payment was enabled — autonomous triggers now correctly bill the agent's team, matching the behavior of other platform channels. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.44...v0.17.45 High4/15/2026
v0.17.44# Release v0.17.44 ## Bug Fixes - Fixed OpenRouter-based image generation for both auto-generated agent/team avatars and image skills — generation now succeeds reliably. ## Improvements - Migrated the OpenRouter integration to the official OpenRouter Python SDK for better retry handling and type safety. - Traffic going through OpenRouter is now attributed to IntentKit in the OpenRouter dashboard (previously attributed to LangChain). - Added cost reconciliation logging so OpenRouter-reported High4/15/2026
v0.17.43# Release v0.17.43 ## New Features - Added a team avatar generation API that lets user and team profiles create AI-generated avatars from a text prompt, with usage billed through a new direct media billing path. ## Improvements - Improved WeChat integration reliability — media uploads to the WeChat CDN now retry automatically on transient failures. - Internal refresh of the Go integration documentation and dependency upgrades. **Full Changelog**: https://github.com/crestalnetwork/intentkit/High4/14/2026
v0.17.42# Release v0.17.42 ## New Features - Sub-agent calls (via `call_agent`) now surface attachments (images, files, etc.) from the sub-agent back to the calling agent. ## Bug Fixes - Fixed a critical issue in the skill pricing module where skill calls were not being charged their declared per-skill price. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.41...v0.17.42 High4/14/2026
v0.17.41# Release v0.17.41 ## Bug Fixes - Fixed a bug where sub-agent calls (via `call_agent`) reused the same conversation thread across invocations, causing stale message history to interfere with new skill configurations. Sub-agent calls are now stateless — each call gets a fresh conversation context. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.40...v0.17.41 Medium4/13/2026
v0.17.40# Release v0.17.40 ## Bug Fixes - Fixed WeChat image sending — images now display correctly instead of falling back to text links - Rewrote WeChat CDN upload protocol to correctly implement the iLink API specification (client-side encryption key generation, proper upload flow, correct message format) **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.39...v0.17.40 Medium4/13/2026
v0.17.39# Release v0.17.39 ## Bug Fixes - Fixed image and video files not displaying correctly in WeChat and mobile browsers due to missing file extensions - Fixed Telegram messaging issues **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.38...v0.17.39 Medium4/13/2026
v0.17.38# Release v0.17.38 ## Improvements - Improved type safety across the codebase, resolving 122 type checker errors in source code, models, and tests - Fixed Pydantic model field defaults to be compatible with static type checkers ## Bug Fixes - Fixed a model name typo **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.37...v0.17.38 Medium4/13/2026
v0.17.37# Release v0.17.37 ## New Features - **Skill Validation for Agent Creation**: The agent manager now validates skill configurations when creating new agents, ensuring only valid skills can be assigned. Invalid skill names or formats are rejected with clear error messages. - **Enhanced Skill Listing**: The available skills listing now shows individual skill names and descriptions under each category, making it easier for the agent manager to select the right skills. - **Improved Agent Manager PrMedium4/13/2026
v0.17.36# Release v0.17.36 ## Bug Fixes - Fixed a startup crash in public agents sync caused by accessing database objects after the session was closed. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.35...v0.17.36 Medium4/13/2026
v0.17.35# Release v0.17.35 ## Bug Fixes - Fixed an issue where the lead agent could not delegate tasks to team agents via WeChat and Telegram channels. The agent routing was using an incorrect identifier, causing sub-agent calls to always fail from these channels. - Added a safety mechanism to prevent Gemini 3 model calls from failing with "empty parts" errors due to corrupted conversation history. ## Improvements - Public agent sync now uses slug-based matching for more reliable updates, and supporMedium4/13/2026
v0.17.34# Release v0.17.34 ## Bug Fixes - Fixed an issue where public agents could not be accessed for chat or content viewing through the Team API. Chat history, messages, activities, and posts for public agents were inaccessible to users outside the agent's owning team. ## Improvements - Improved chat data isolation so that teams can only view and interact with their own conversations on public agents. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.33...v0.17.34 Medium4/13/2026
v0.17.33## Bug Fixes - Markdown images in chat messages are now constrained to a reasonable max height (320px) instead of rendering at full size. - Image attachment thumbnail hover overlay now matches the actual image bounds for portrait/narrow images instead of extending beyond the image area. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.32...v0.17.33Medium4/12/2026
v0.17.32# Release v0.17.32 ## Bug Fixes - Gemini image and Veo video skills now respect Vertex AI configuration (`GOOGLE_GENAI_USE_VERTEXAI`, `GOOGLE_CLOUD_PROJECT`), fixing failures when using Vertex AI credentials instead of a direct API key. - LLM tool selector middleware now uses a dedicated model picker (`pick_tool_selector_model`) restricted to OpenAI, avoiding structured-output incompatibilities with Gemini and GLM that caused tool selection to fail silently or return descriptions instead of toMedium4/12/2026
v0.17.31# Release v0.17.31 ## Bug Fixes - Public agent pages can now be opened directly from the Discover page again, for both signed-in members of other teams and anonymous visitors. **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.30...v0.17.31 High4/10/2026
v0.17.30# Release v0.17.30 ## New Features - Activity push notifications on WeChat and Telegram now include a direct link back to the original post, so recipients can jump straight to the source. ## Improvements - The team lead agent has a clearer role definition and a refined decision-making workflow, helping it more reliably choose between answering directly, delegating to a built-in helper, calling an existing team agent, or creating a new specialized agent when needed. **Full Changelog**: httpsMedium4/10/2026
v0.17.29# Release v0.17.29 ## Improvements - Free credit refills now happen once a day instead of every hour, giving a clearer daily allowance experience - Upgraded the web frontend to the latest Tailwind CSS v4 for a smoother styling pipeline ## Bug Fixes - Fixed an issue where updates to the team lead agent's long-term memory would not take effect until the next cache refresh - Fixed a compatibility issue with older team plan data that could cause errors when loading teams **Full Changelog**: httMedium4/10/2026
v0.17.28# Release v0.17.28 ## New Features - Added a read-only default channel conversation view in the lead agent sidebar, allowing you to browse message history from the default channel (Telegram/WeChat) directly in the web UI ## Improvements - Improved channel-related API path naming for consistency - Added proper WeChat type support in the frontend **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.27...v0.17.28 Medium4/9/2026
v0.17.27# Release v0.17.27 ## New Features - Redesigned the team lead agent with a coordinator and sub-agents architecture, improving task delegation and specialization - Added self-updater and content-manager sub-agents, allowing the team lead to better manage its own configuration and content workflows **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.26...v0.17.27 Medium4/9/2026
v0.17.26## New Features - Added ACP (Agentic Commerce Protocol) skill category, enabling agents to browse and purchase products from ACP merchants using x402 crypto payments ## Bug Fixes - Fixed form validation errors for optional fields with enum constraints in the agent creation form **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.25...v0.17.26 Medium4/9/2026
v0.17.25## Improvements - Avatar images are now automatically normalized to a consistent 512x512 square format before uploading to CDN, ensuring uniform display across all platforms - Updated various dependencies to latest versions **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.24...v0.17.25 High4/9/2026
v0.17.24## Bug Fixes - Fixed an issue where the team reward script failed to run due to uninitialized database connection **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.23...v0.17.24 Medium4/8/2026
v0.17.23## New Features - OpenRouter agents now use native server tools for web search and datetime, improving response speed and accuracy - Added new LLM models: Gemma 4 31B, upgraded GLM 5.1, and updated Qwen 3.6 Plus pricing ## Improvements - Streamlined internal system skills architecture for cleaner conditional loading - Various bug fixes and test improvements **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.22...v0.17.23 Medium4/8/2026
v0.17.22## New Features - Users can now leave a team from their account page (owners must transfer ownership first) - Teams now include plan details (name, description, seats, pricing) for frontend display - Team list API now returns each user's role for proper role-based UI ## Improvements - Improved EVM wallet detection for users who registered via Web3 login - Simplified account linking by removing unused EVM wallet unlink flow - Enriched team members API with additional user details **Full ChangMedium4/8/2026
v0.17.21- Added push channel system for team lead agents, enabling proactive notifications - Adjusted account monitoring schedule for improved efficiency **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.20...v0.17.21 Medium4/7/2026
v0.17.20- Migrated Z.AI web search and webpage reader from REST API to MCP protocol for improved reliability - Optimized LLM model selection priority for better performance and cost efficiency - Fixed an issue with team plan handling that could cause errors in certain configurations - Improved conversation cancellation reliability **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.19...v0.17.20 Medium4/4/2026
v0.17.19# v0.17.19 - Provider-based web search: web search and webpage reading now use provider-native capabilities (OpenAI, xAI, Google, OpenRouter), with Z.AI fallback for other providers - Added Z.AI web search and webpage reader skills - Consolidated search logic from middleware into executor for cleaner architecture **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.18...v0.17.19 Medium4/4/2026
v0.17.18# v0.17.18 - Added Qwen 3.6 Plus model support - Fixed an issue where team agent chat was not working due to missing billing context - Updated dependencies **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.17...v0.17.18 Medium4/3/2026
v0.17.17# v0.17.17 - Added Anthropic Compatible LLM provider, allowing connection to any Anthropic-API-compatible service via environment variables **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.16...v0.17.17 Medium4/3/2026
v0.17.16# v0.17.16 - Fixed a crash on the agent chat page that could occur when loading agents **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.15...v0.17.16 Medium4/3/2026
v0.17.15# v0.17.15 - Shared schema endpoint between local and team APIs (moved to app/common) - Extracted RJSF form utilities (validator, uiSchema, cleanSkillsData) into shared formUtils.ts - Added AI avatar hint to PictureWidget - Cleaned up duplicate code between agent create/edit pages **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.14...v0.17.15Medium4/3/2026
v0.17.14# v0.17.14 - Added team usage page API for viewing credit balances and recent activity - Updated billing roadmap documentation to reflect completed milestones **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.13...v0.17.14 Medium4/3/2026
v0.17.13# v0.17.13 - Added user profile with display name and avatar support - Users can now edit their name and upload an avatar in account settings - Google sign-up automatically populates name and avatar from Google account - EVM wallet sign-up generates a display name from the wallet address - Linking a Google account syncs avatar if not already set - Fixed tasks page header alignment and skeleton persistence during navigation - Improved team ID validation for the API layer - Fixed Google image modMedium4/2/2026
v0.17.12# v0.17.12 - Fixed authentication failure caused by missing JWT audience validation **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.11...v0.17.12 Medium4/2/2026
v0.17.11# v0.17.11 - Fixed authentication failure in Team API where login always returned 401 due to JWT algorithm mismatch with modern Supabase signing keys **Full Changelog**: https://github.com/crestalnetwork/intentkit/compare/v0.17.10...v0.17.11 Medium4/2/2026

Dependencies & License Audit

Loading dependencies...

Similar Packages

opentulpaSelf-hosted personal AI agent that lives in your DMs. Describe any workflow: triage Gmail, pull a Giphy feed, build a Slack bot, monitor markets. It writes the code, runs it, schedules it, and saves imain@2026-06-05
SimpleLLMFuncA simple and well-tailored LLM application framework that enables you to seamlessly integrate LLM capabilities in the most "Code-Centric" manner. LLM As Function, Prompt As Code. 一ä¸ĒįŽ€å•įš„æ°åˆ°v0.8.4
deer-flowAn open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tamain@2026-06-06
ag2AG2 (formerly AutoGen): The Open-Source AgentOS.Join us at: https://discord.gg/sNGSwQME3xv0.13.3
agent-frameworkA framework for building, orchestrating and deploying AI agents and multi-agent workflows with support for Python and .NET.dotnet-1.9.0

More in Frameworks

langchainThe agent engineering platform
deer-flowAn open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of ta
tqdmFast, Extensible Progress Meter
simBuild, deploy, and orchestrate AI agents. Sim is the central intelligence layer for your AI workforce.