Thunderbolt
The AI you control: choose your models, own your data, and eliminate vendor lock-in.
- Source repo
- thunderbird/thunderbolt
- Stars
- ★ 4.8k
- Last updated
- 1d ago
- License
- MPL-2.0
- Primary language
- TypeScript
- FA score
- 49/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- An enterprise IT team wants to deploy an AI chat tool on-premises to meet data privacy and compliance requirements.
- Not a fit if
- Individuals wanting a ready-to-use app with no backend setup
- Enterprises needing production-ready software; still under security audit
- Source review
- 49/100 · Major gaps
What does this agent do, and when should you use it?
Thunderbolt is an open-source, cross-platform AI client developed by the Thunderbird team, designed for on-premises deployment in enterprise environments. It supports all major desktop and mobile platforms, including web, iOS, Android, Mac, Linux, and Windows. Thunderbolt is compatible with frontier, local, and on-prem models, and offers enterprise features, support, and FDEs. The project is under active development, currently undergoing a security audit, and preparing for enterprise production readiness. It provides a self-hosted backend with Docker Compose or Kubernetes, using Postgres and PowerSync for data management. Users must bring their own model providers; there is no public inference endpoint, but you can use Ollama or llama.cpp for free local inference or add API keys for any OpenAI-compatible model provider. Thunderbolt is licensed under MPL-2.0 and adheres to Mozilla's Community Participation Guidelines.
Thunderbolt provides a unified chat interface to interact with multiple AI model providers. It runs a backend service (port 8000) for authentication and data sync, and a frontend (port 1420) built with Vite and Tauri. Users can configure API keys for OpenAI-compatible providers or local inference servers like Ollama or llama.cpp. The app supports disabling search functionality to move toward offline-first operation. The development workflow uses Make commands: make doctor verifies tooling, make setup installs dependencies and symlinks agents, make up starts Postgres and PowerSync in Docker, and make run launches both backend and frontend. The repository includes architecture docs, a Storybook for component development, and a Vite Bundle Analyzer for performance tuning.
- An enterprise IT team wants to deploy an AI chat tool on-premises to meet data privacy and compliance requirements.
- A developer wants to use local open-source models via Ollama or llama.cpp and interact through a polished chat interface.
- An organization needs a cross-platform AI client (desktop and mobile) with a self-hosted backend to avoid cloud dependency.
- A power user wants to switch between multiple model providers without being locked into a single vendor.
- A team wants to evaluate AI productivity tools while maintaining full control over their data and infrastructure.
How do you install or deploy this agent?
Thunderbolt is still early-stage and does not yet provide prebuilt installers for end users. Developers can set up a local environment by cloning the repository and running: make doctor to check tools, make setup to install dependencies and link agent symlinks, make up to start Postgres and PowerSync in Docker, and make run to start the backend (port 8000) and frontend (port 1420). For enterprise self-hosting, refer to the deploy/README.md for Docker Compose or Kubernetes configurations. You will need to configure your own model provider, such as installing Ollama or llama.cpp.
How do you use this agent?
After installation, open the frontend (in development, http://localhost:1420), register or sign in (requires self-hosted backend). In the settings, add your model provider: either an API key for an OpenAI-compatible service or the endpoint for a local Ollama/llama.cpp instance. Select a model from the chat interface and start chatting. To move closer to offline usage, disable search in the integrations settings.
What are this agent's strengths and limitations?
- Fully open-source (MPL-2.0) and self-hostable, giving complete control over data and deployment.
- Cross-platform support across web, iOS, Android, Mac, Linux, and Windows.
- Model-agnostic, compatible with any OpenAI-compatible API, and encourages local inference with Ollama or llama.cpp.
- Enterprise features, support, and FDEs available, making it suitable for production environments.
- Strong emphasis on security (security audit) and community participation guidelines.
- Still in early development; not fully offline-first, currently depends on authentication and search (search can be disabled).
- No public inference endpoint, so users must configure their own model providers, increasing setup complexity.
- Targeted primarily at enterprise customers; individual users may find documentation and installation challenging.
- Relies on external services (Postgres, PowerSync), adding deployment and operational overhead.
- Currently undergoing security audit, so production readiness is not yet fully established.
How does this agent compare with similar options?
Thunderbolt differentiates from closed-source clients like OpenAI's ChatGPT or Anthropic's Claude by emphasizing data ownership and model freedom, allowing full self-hosting and avoiding vendor lock-in. Compared to other open-source chat clients, Thunderbolt focuses on enterprise-grade deployment and model neutrality.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Thunderbolt This agent | 49 · Major gaps | Desktop appFree + model costs | ★ 4.8k | 1d ago | TypeScript | OpenAI API |
| little-coder: Lightweight Local Coding Agent | 58 · Major gaps | CLIFree + model costs | ★ 2.7k | 1mo ago | TypeScript | — |
| Forge: Reliability Layer for Self-Hosted LLM Tool Calling | 49 · Major gaps | CLIFree + model costs | ★ 2.3k | 4d ago | Python | OpenAI API · Claude API |
| AI Town | 47 · Major gaps | CLIFree + model costs | ★ 11k | 1mo ago | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows security audit claims, dependency locking, and CI security gates, but lacks specific implementation details for least privilege, user confirmation, data flow transparency, and sensitive data handling. Deductions: no clear evidence of least privilege, user confirmation, data flow diagrams, or sensitive data encryption.
Project has CI tests and dependency locking, but failure messages and error handling documentation are insufficient. Deductions: specific examples of failure messages and error recovery mechanisms are not documented.
Project targets enterprise users, supports multiple platforms and models, but capability boundaries and trigger precision are unclear. Deductions: no clear description of agent capability boundaries and trigger conditions.
Project has clear documentation structure, installation guides, and FAQ, but version changelog and naming stability information are insufficient. Deductions: missing CHANGELOG and naming stability statements.
Project offers local deployment and model choice, but specific evidence of output usability and marginal value is insufficient. Deductions: no output format examples or real-world use cases.
Project has CI and tests, but the correspondence between claims and evidence is unclear. Deductions: no traceable claim sources or cross-validation.
- Project is early-stage, depends on external authentication and search services, may not be suitable for fully offline environments.
- Publisher identity is unverified, supply chain risks should be carefully assessed.
- Security audit is not yet complete, production use should wait for audit results.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.1.144 Release v0.1.144
Release notes excerpt
Release v0.1.144 Warning: This is an automated build and may be unstable. Use at your own risk. - Version: 0.1.144 - Version type: patch Changelog [0.1.144] - 2026-10-17 Features - Treat / as plain text in minimal mode (#1403) (d4929ba) - Require a step-up OTP to delete an account (#1356) (ea0f619) Fixes - Cut a revoked device's sync stream and wipe it (#1369) (4cdf4f8) Platform Builds - ✅ Desktop: Building via CI - ✅ iOS: Building and deploying to TestFlight - ✅ Android: Building and deploying to Play Store (Internal)View on GitHub ↗ -
v0.1.143 Release v0.1.143
Release notes excerpt
Release v0.1.143 Warning: This is an automated build and may be unstable. Use at your own risk. - Version: 0.1.143 - Version type: patch Changelog [0.1.143] - 2026-10-17 Features - Native passkeys on Android (#1401) (4863a0e) Platform Builds - ✅ Desktop: Building via CI - ✅ iOS: Building and deploying to TestFlight - ✅ Android: Building and deploying to Play Store (Internal)View on GitHub ↗ -
v0.1.142 Release v0.1.142
Release notes excerpt
Release v0.1.142 Warning: This is an automated build and may be unstable. Use at your own risk. - Version: 0.1.142 - Version type: patch Changelog [0.1.142] - 2026-10-17 Features - Minimal mode for public deployments (#1402) (43bf9fe) - Let the hosted agent call read-only MCP tools (#1394) (6585c52) Platform Builds - ✅ Desktop: Building via CI - ✅ iOS: Building and deploying to TestFlight - ✅ Android: Building and deploying to Play Store (Internal)View on GitHub ↗