awesome-repositories.com
Blog
MCP
awesome-repositories.com

Discover the best open-source repositories with AI-powered search.

ExploreCurated searchesOpen-source alternativesSelf-hosted softwareBlogSitemap
ProjectMCP serverAboutHow we rankPress
LegalPrivacyTerms
© 2026 Bringes Technology SRL·VAT RO45896025·hello@awesome-repositories.com
thunderbird avatar

thunderbird/thunderbolt

0
View on GitHub↗
thunderbolt.io↗

Thunderbolt

Thunderbolt is an open-source, self-hostable AI chat client that runs on web, desktop (macOS, Linux, Windows) and mobile (iOS, Android) from one shared web codebase inside lightweight native shells. It is local-first: each device keeps a local database as the source of truth, and data syncs to a central backend with optional end-to-end encryption, so the server stores only ciphertext for protected content. New devices are approved from trusted ones by fingerprint, recovery phrase or passkey, and revoked devices lose access.

What sets it apart is its approach to model and agent choice and to privacy. Users can connect hosted providers, any OpenAI-compatible endpoint, local models, or a managed catalog with per-account spend limits. A confidential inference tier encrypts requests for a hardware-attested enclave, so the relaying server cannot read the conversation. Outside coding and search agents can be attached to any chat over open agent protocols, over WebSockets or encrypted peer-to-peer links. Tool servers connected through the Model Context Protocol are merged into the model's toolset, and a terminal coding agent and a browser-based coding agent are also included.

Around the core chat, it offers projects with shared instructions, reusable skills, file attachments, voice conversations, search across local data, and rich in-chat widgets. Model-written HTML pages render in a sandboxed frame, and embedded mini apps can share context with the chat. Email, calendar and file integrations are available for Google and Microsoft accounts.

On the server side, sign-in works through OIDC, SAML, emailed codes or passkeys. The server stack deploys with Docker Compose, Kubernetes with Helm, or AWS infrastructure-as-code. It also covers rate limits, usage metering, and optional telemetry that is off by default.

Features

  • AI Chat Clients - Provides a self-hostable, local-first chat application for talking to AI models of your choice on web, desktop, and mobile.
  • AI Chat Clients - Delivers a desktop, web and mobile chat client for talking to AI models, organizing conversations into chats and projects.
  • Agent Client Protocol Connections - Connects outside AI agents to chats through the Agent Client Protocol, with settings to configure and manage each connected agent.
  • Bring-Your-Own-Key Model Clients - Connects to hosted providers, any OpenAI-compatible endpoint, confidential enclaves, or local models using personal keys or a managed catalog with spend limits.

AI search

Explore more awesome repositories

Describe what you need in plain English — the AI ranks thousands of curated open-source projects by relevance.

Start searching with AI
4,819 stars·338 forks·TypeScript·MPL-2.0·1 view
  • Bring-Your-Own-Key Model Access - Connects hosted providers with the user's own API key or any OpenAI-compatible endpoint by URL so users choose which model answers.
  • External Agent Integrations - Attaches outside coding and search agents to any chat over open agent protocols and merges tool servers into the model's toolset.
  • Model Provider Management - Connects chats to hosted providers, any compatible endpoint, or local models, and can offer a managed catalog with per-account spend limits.
  • Managed Model Catalogs with Spend Limits - Offers deployment-supplied models that work without keys, metered per account against rolling five-hour and seven-day spending allowances.
  • Private AI Chat Deployments - Deploys the full chat service on your own servers, Kubernetes cluster, or AWS account so conversations, models, and user data stay in your control.
  • Local-First Sync Libraries - Keeps a local database on each device as the source of truth, reading and writing locally first, with optional sync across devices.
  • Local-First Database Replication - Replicates each user's data between a central PostgreSQL backend and local SQLite on clients, scoped by a signed identity token.
  • Encrypted Sync Clients - Stores data locally first and syncs it across devices with optional end-to-end encryption so only the user can read it.
  • Self-Hosted Deployments - Runs the backend, database, sync service and identity provider on own infrastructure using containers, Kubernetes clusters or infrastructure-as-code.
  • Multi-Platform App Delivery - Delivers the same AI chat client on web, iOS, Android, macOS, Linux and Windows from one shared codebase.
  • AI Model Clients - Provides one AI chat client running on web, iOS, Android, Mac, Linux and Windows so the assistant is the same everywhere.
  • Data Protection - Applies encryption and device attestation so user data stays under the user's control rather than a single vendor.
  • Encrypted Multi-Device Chat Sync - Keeps chats, settings and projects on every device with local-first storage, optional end-to-end encryption, and device approval and revocation.
  • End-to-End Encryption - Applies optional end-to-end encryption so the server stores only ciphertext and cannot read chat histories.
  • Confidential Inference Enclaves - Verifies hardware attestation of the confidential inference connection so users can confirm requests reach a trusted enclave the relaying server cannot read.
  • Single Sign-On Integrations - Signs users in through existing OIDC or SAML identity providers, emailed codes or passkeys so access follows organizational identity rules.
  • Local-First Architectures - Keeps a local SQLite database on each device as the source of truth and replicates changes with a central PostgreSQL backend.
  • Cross-Platform Desktop Shells - Runs one shared web interface inside lightweight native windows on desktop and mobile, with thin platform-specific adapters around it.
  • Native Shell Embeddings - Runs one shared React codebase unchanged in the browser and inside lightweight native windows on desktop and mobile with thin platform adapters.
  • Slash-Inserted Text Skills - Provides reusable named instruction blocks applied by slash command or loaded automatically by the model when a request matches, pinned above the composer.
  • Agent Tool Server Connections - Gives the hosted agent read-only tools from configured tool servers, with server-held credentials, tool and call limits, and graceful handling of unreachable servers.
  • MCP Server Connections - Adds Model Context Protocol servers through settings and merges their tools into the model's toolset on every message.
  • Streaming Chat Routing - Forwards chat requests to managed inference backends and streams replies back, applying usage metering and per-user limits.
  • External Agent Delegations - Hands a conversation to an external coding agent over a network connection, with a terminal agent available for command-line work.
  • Remote Agent Connections - Lets any chat point at an agent running on a server or laptop via a WebSocket address or peer-to-peer ticket, with its commands added to the slash menu.
  • Voice-to-Voice Loops - Provides voice mode with speech-to-text, text-to-speech, mic-open cues and acoustic barge-in through hosted or custom providers.
  • AI Provider Integrations - Adds models from several hosted providers or any compatible endpoint, called directly from the device or through an optional backend proxy.
  • Browser-Based Coding Agents - Includes a full coding agent that runs inside the browser tab with shell and file tools over a virtual filesystem and no server execution.
  • Conversational Voice AI - Captures microphone speech, transcribes each utterance, routes it through normal chat, and reads the reply aloud hands-free.
  • Automatic Context Compaction - Summarizes older conversation history manually or automatically near the context limit while keeping recent turns, so long chats continue without overflowing the model window.
  • Model Catalog Management - Offers confidential and managed inference providers, swaps built-in models via a skill, and requires a successful connection test before saving a custom model.
  • Runtime Provider Switching - Switches among many model providers or any OpenAI-compatible endpoint, with saved per-profile credentials and per-run provider, model and reasoning flags.
  • Model Provider Freedom - Works with frontier, local, and on-prem models so users pick their own model and avoid lock-in to a single provider.
  • Multi-Backend Agent Chat - Points any chat at an outside agent over an open protocol or the built-in assistant, with agents added by the backend or users.
  • Reusable Instruction Skills - Stores named reusable instruction blocks in a synced library users create, edit, enable, disable and delete from settings.
  • Chat Workflow Customization - Lets users shape assistants for repeated work with reusable skills, shared project instructions, embedded mini apps, and interactive widgets inside chats.
  • Terminal Coding Agents - Operates in the working directory with shell, file read/write/edit and web fetch tools, taking a single prompt or interactive session from one binary.
  • Speech-to-Speech Frameworks - Transcribes speech and reads replies aloud via a hosted enclave speech engine or any OpenAI-compatible speech endpoint configured per device.
  • Chat Project Organization - Groups related chats into projects whose shared instructions are inherited by every conversation inside them.
  • Shared Project Instructions - Stores durable instructions on a project and injects them into the system prompt of every chat inside it for consistent guidance.
  • Cross-Device Chat History Syncs - Keeps conversations in a local-first database and syncs them between devices with opt-in end-to-end encryption.
  • Local Data Engines - Keeps a local SQLite database as each device's source of truth so the app stays fully usable offline.
  • Offline Write Queues - Saves edits to the on-device database immediately, queues them while disconnected, and replays them on reconnect with latest-write-wins conflict handling.
  • Agent Protocol Bridges - Bridges external coding agents into chats over a standard agent protocol on WebSockets or peer-to-peer links, mapping slash commands and sessions.
  • Tool Server Merging - Connects Model Context Protocol tool servers over several transports and merges their namespaced tools into the model's toolset on every message.
  • Confidential Inference Attestation Checks - Verifies hardware attestation, signed build provenance, measurements and key binding on the client and shows pass or fail status in the model selector.
  • Cross-Device Document Syncs - Streams changes between a local database on each device and a backend database so chats, settings and configurations stay consistent.
  • AWS ECS Deployments - Provisions AWS network and compute as code with Pulumi, targeting serverless ECS Fargate containers or an EKS cluster running the Helm chart.
  • On-Premise Deployment - Installs on infrastructure the organization controls so conversations and data stay within its own environment rather than a vendor cloud.
  • Self-Hosted Agent Inference - Enables direct, confidential and hosted-agent inference tiers on a private deployment through environment variables for provider keys, enclave address and quota budgets.
  • Cross-Platform Desktop and Mobile Apps - Runs one shared React codebase inside native shells on macOS, Linux, Windows, iOS and Android so every platform gets the same experience.
  • Email and Calendar Integrations - Connects Gmail, Google Calendar, Outlook and OneDrive so the assistant can read, search and draft mail and work with events and files.
  • Chat History Encryption - Encrypts synced chat histories so the server cannot read them, with optional key escrow for enterprise deployments.
  • Confidential Request Relaying - Passes end-to-end encrypted requests untouched through the backend to a hardware-isolated enclave, stripping credentials and unsafe headers so session secrets never reach the model host.
  • Synced Column Decryption - Intercepts incoming sync data and decrypts protected columns with AES-256-GCM before local storage, so the server only ever holds ciphertext.
  • Fingerprint-Verified Device Approvals - Adds new devices by having the user verify a 256-bit key fingerprint on a trusted device before sharing the account key.
  • Device Revocation and Key Rotation - Cuts a lost device's API and sync access and rotates the account and data keys so it can never read new data.
  • Enterprise Identity Providers - Redirects users to a configured OIDC or SAML identity provider and links returning identities to existing accounts safely.
  • Identity Providers - Ships a bundled Keycloak identity provider with a demo realm and also works with any external OIDC or SAML provider.
  • Enterprise SSO Authentication - Redirects unauthenticated users through the organization's identity provider, exchanges the returned auth code for tokens and creates or updates the session.
  • SAML Integrations - Connects the deployment to OIDC or SAML identity providers such as Keycloak, Okta, Auth0 or Entra ID for enterprise sign-in.
  • Passkey Authentication - Lets users sign in with passkeys through WebAuthn and unlock sync on a new device, with configurable relying party identity.
  • Post-Quantum Key Exchange - Wraps each device's account key using classical elliptic-curve and lattice-based post-quantum key exchange together, so both must be broken.
  • Database Column Encryption - Encrypts columns listed in a central configuration map before upload, so the server only receives ciphertext for those fields.
  • Layered Key Hierarchies - Uses an account key that wraps versioned data keys and is sealed to each approved device, so the server stores only ciphertext.
  • Agent Tool Permissions - Asks users to allow or deny external agent tool requests, remembers prior allowances per agent and tool kind, and cancels pending requests on termination.
  • Single Sign-On - Supports browser-redirect single sign-on including a desktop initiate-and-callback flow, with OIDC, Google and Microsoft sign-in, token exchange and refresh.
  • Single Sign-On Providers - Authenticates users through the organization's own single sign-on provider, falling back to emailed sign-in codes when none is available.
  • Trusted Device Management - Shows live lists of trusted devices and pending approvals, keeps revoked devices visible briefly, and offers pairing details with a QR code.
  • Assistant Skill Plugins - Adds reusable skills, mini-apps, widgets and task extensions that expand what the assistant can do beyond plain conversation.
  • Usage Limiters - Enforces authentication, device binding, rate limits and rolling-window spending budgets on every managed inference route.
  • Chat-Embedded Widgets - Parses widget tags in model replies and maps each to a component, showing forecasts, maps and link previews inline in chat.
  • Sandboxed HTML Rendering - Runs model-written HTML pages in one shared sandboxed frame after a self-containment check, with a scripts-off live preview while streaming.
  • Chat File Attachments - Adds files to a chat turn for the model to use while keeping file contents on the device rather than the server.
  • Mini App Runtimes - Runs ordinary web apps inside an iframe in the client so they can publish context to the chat and receive actions from it.
  • Sandboxed Generated Page Rendering - Runs model-written self-contained HTML pages in a single sandboxed frame so generated code executes safely on the user's device.

    • Star history

    Star history chart for thunderbird/thunderboltStar history chart for thunderbird/thunderbolt

    How this analysis was created: This summary and feature list are AI-generated from collected project material and can contain mistakes. Stars, license and language are imported from GitHub. Inclusion does not mean that we have tested or audited this project. Check the source documentation for any feature you depend on. Learn more on our About page.

    Frequently asked questions

    What does thunderbird/thunderbolt do?

    Thunderbolt is an open-source, self-hostable AI chat client that runs on web, desktop (macOS, Linux, Windows) and mobile (iOS, Android) from one shared web codebase inside lightweight native shells. It is local-first: each device keeps a local database as the source of truth, and data syncs to a central backend with optional end-to-end encryption, so the server stores only ciphertext for protected content. New devices are approved from trusted ones by fingerprint, recovery…

    What are the main features of thunderbird/thunderbolt?

    The main features of thunderbird/thunderbolt are: AI Chat Clients, Agent Client Protocol Connections, Bring-Your-Own-Key Model Clients, Bring-Your-Own-Key Model Access, External Agent Integrations, Model Provider Management, Managed Model Catalogs with Spend Limits, Private AI Chat Deployments.

    Which projects share features with thunderbird/thunderbolt?

    Projects with overlapping indexed features include: genspark-ai/genoffice — GenOffice is an open-source desktop office suite with editors for Word, Excel and PowerPoint style files, plus PDF,… elie222/rakazo — Rakazo is a self-hosted, open-source platform of persistent AI bots. Each bot keeps its own conversations, memory,… mnfst/llm-gateway — LLM Gateway is an open-source, self-hostable gateway that sits between applications and AI model providers. It accepts… agenticresourceregistry/agenticresourceregistry — Agentic Resource Registry is a self-hosted catalog where teams publish, version, review and discover reusable coding… snouzy/workout-cool — Workout-Cool is a self-hosted fitness application that combines an exercise database browser, workout plan builder,… automerge/automerge-classic — Automerge Classic is a JavaScript and TypeScript library for building collaborative, local-first applications. It is…

    Projects sharing features with Thunderbolt

    These projects share indexed features with Thunderbolt. Shared tags can include platform or build tooling; verify the primary use case before treating a result as a replacement.
    • genspark-ai/genofficegenspark-ai avatar

      genspark-ai/genoffice

      11,901View on GitHub↗

      GenOffice is an open-source desktop office suite with editors for Word, Excel and PowerPoint style files, plus PDF, Markdown and HTML, and a built-in AI agent that drafts and revises documents. Files are opened, edited, converted and searched on the user's own machine, and a network connection is needed only for the AI features, which run on the user's own model keys. Several things set it apart. Edits are applied as narrow patches that leave every untouched part of the original file byte-for-byte intact, so documents stay compatible with Microsoft Office. In Word files, AI rewrites appear as

      TypeScriptaiai-agentclaude-code
      View on GitHub↗11,901
    • elie222/rakazoelie222 avatar

      elie222/rakazo

      7,458View on GitHub↗

      Rakazo is a self-hosted, open-source platform of persistent AI bots. Each bot keeps its own conversations, memory, scheduled routines, and history, and can hand work to peer bots or short-lived helper agents. Bots get isolated shared or private computers with a browser, terminal, files, and a graphical desktop, and users can watch a bot's screen and take control of it when needed. The same bots are reachable from web, desktop, and mobile apps, and from messaging channels such as iMessage, Slack, Telegram, and WhatsApp. The model and the computer are both your choice. You can connect your own

      TypeScriptai-agentschatgptdocker
      View on GitHub↗7,458
    • mnfst/llm-gatewaymnfst avatar

      mnfst/llm-gateway

      7,625View on GitHub↗

      LLM Gateway is an open-source, self-hostable gateway that sits between applications and AI model providers. It accepts requests at a single OpenAI-compatible chat completions endpoint, so any SDK that allows a custom base URL can send traffic through it, and forwards each request to a suitable model from the connected providers. Each agent uses its own key, and a PostgreSQL database backs the service. What sets it apart is the routing and cost layer. It can score each request and send it to the cheapest model able to handle it, pick a model automatically when the model is set to auto, or appl

      TypeScriptaiai-ai-sdk
      View on GitHub↗7,625
    • agenticresourceregistry/agenticresourceregistryAgenticResourceRegistry avatar

      AgenticResourceRegistry/AgenticResourceRegistry

      4,325View on GitHub↗

      Agentic Resource Registry is a self-hosted catalog where teams publish, version, review and discover reusable coding agent resources, including agents, skills, tool servers, hooks, prompts and sandboxes. It also stores and serves Model Context Protocol tool servers alongside the other resource types. Organizations run it on their own infrastructure, so shared resources and audit data stay under internal control. Submitted versions go through a review process in which reviewers inspect redacted files and differences, discuss changes, and approve or request edits. Publishing happens only after

      Pythonagentsanalyticsantigravity
      View on GitHub↗4,325
    Compare all 30 related projects→