gemma3:1b), an open-weight, lightweight language model capable of running smoothly on laptops with minimal RAM and CPU/GPU resources.http://localhost:11434), serving local chat completions via standard REST endpoints.Correction / Why / Continue blocks, and refrain from critiquing or hallucinating errors.window.speechSynthesis API (0 external dependencies) so learners can listen to the correct pronunciation at a comfortable tempo.
Check out the video walkthrough of SpeakMate in action, showing live conversational practice, gentle grammar correction cards, and local inference:
🎥 Watch the Video Demo on YouTube The complete source code is open source under the MIT License on GitHub:
Practice English naturally with your private AI conversation partner.
Built as an open-source Hacktoberfest 2026 project.
SpeakMate is a friendly, distraction-free AI English conversation partner designed for learners who want to practice conversational English without pressure. SpeakMate engages in natural dialog while gently offering clear grammar and wording corrections with simple explanations.
Best of all, SpeakMate runs 100% locally on your machine using Ollama and Gemma 3 1B. No OpenAI, Gemini, or Claude API keys required. No cloud subscriptions. Zero telemetry. Complete privacy.
Practicing a new language with a native speaker can sometimes feel intimidating or stressful. Traditional grammar check tools are rigid and feel like spellcheckers, while cloud LLMs can be costly, require API keys, and send personal conversations to third-party servers.
SpeakMate was built to solve this:
🔗 Repository: https://github.com/Ahzem/speakmate Why build with open-source AI and local models instead of calling a closed cloud API? For an application like SpeakMate, open innovation is not just an implementation detail — it's the core reason the product works:
Psychological Safety Through Total Privacy:
Practicing a language means making mistakes. When users know that every sentence is processed entirely on their laptop's RAM and never sent to a cloud server, their anxiety drops. Open weights give users 100% data sovereignty.
Zero Operating Cost:
Cloud APIs charge per token and require credit cards or monthly subscriptions. By running an open-weight 1B parameter model locally, SpeakMate costs $0 to operate. My friend can practice for hours every single day without worrying about API limits or surprise bills.
Offline Resilience:
My friend can practice on the train, in a coffee shop with spotty Wi-Fi, or on a flight. Because Gemma 3 1B runs on-device through Ollama, SpeakMate works completely offline.
Freedom from Vendor Lock-In & Deprecation:
Closed APIs regularly deprecate model versions, alter system guardrails, or change pricing tiers. With open models and open-source tooling, this software will continue to run exactly as intended for years to come.
This project was developed with the assistance of agentic pair-programming in Google Antigravity IDE, configuring the local Ollama integration, structuring React 19 hydration-safe UI components, and tuning Gemma 3 1B's few-shot system prompt.
Thank you to the DEV Community and the Hacktoberfest team for celebrating open innovation and community building!