Redmine AI plugin BYOLM – bring the LLM you already trust.
For Redmine shops – Anthropic, OpenAI, Azure, or local Ollama. From $1.99/mo Cloud.
Why this matters
Atlassian Rovo locks you to one model. Easy8 locks you to one model. We don’t.
Vendor lock-in on the LLM is the new vendor lock-in on the database. Models change pricing. Models change capability. Models go away. Pick the one that fits today and switch when something better lands – without re-platforming the rest of your stack.
Supported providers
Five paths today, Qwen-3.5B in roadmap.
OpenAI
GPT-4o, GPT-4o-mini, future frontier models. Configure once at the AI Connection level. Per-prompt, per-instance, or per-user. API key stays in your Redmine admin config.
Anthropic
Claude Sonnet, Claude Opus. The chatbot’s default for the AI Assistant launch. Best fit for tool-calling and multi-step reasoning over Redmine data.
Azure OpenAI
Same GPT-4o models, deployed to your Azure tenant. Compliance-friendly for organisations with Microsoft contracts. Custom endpoint URL + key, configured per AI Connection.
Ollama (local)
Run Llama 3, Qwen2.5, Mistral, or any GGUF model on your own hardware. Zero data leaves your network. Apple Silicon, NVIDIA T4, or CPU-only – all supported via Ollama’s OpenAI-compatible endpoint.
Qwen-3.5B for Redmineflux
A specialist Qwen2.5-Coder LoRA fine-tune trained on Redmine project tasks – same pipeline we use for Odient (our Odoo equivalent). 1 GB GGUF, runs on a Mac mini. Release tied to data-quality milestones, not a calendar date.
LiteLLM router
For organisations running multiple providers, the chatbot routes through LiteLLM. Pick a model per agent, per task, or per cost ceiling. Fail over to a backup if a provider has an outage. Track spend per model, per user, per project.
Sovereignty by default. Not by upgrade.
Not every question needs Opus 4.7 with a 1-million-token context. Redmineflux routes each query to the model that gives the best cost-and-quality answer for that question – sometimes a local Qwen, sometimes Sonnet, sometimes a frontier model from your provider of choice. You set the policy. The system picks the model. Token spend stays sane.
Choosing your LLM is the easy part. Knowing where the prompt goes is harder. Redmineflux is built on Redmine – open source, self-hostable, your-database-by-default. The AI Assistant plugin honours that. EU and JP data residency are first-class. On-prem Ollama means no prompt content ever leaves your network. Audit logs at the MCP layer record every tool call.
When does each ship?
Honest dates.
Live today
OpenAI / Azure OpenAI / Ollama in the AI plugin. Configure via the admin AI Connection page. Provider-agnostic by design.
Shipping next release
Anthropic Claude Sonnet/Opus integration. Default model for the AI Assistant launch. Configurable per instance.
Shipping next release
LiteLLM routing. Pick provider per task, fail over, track spend.
In trial
Per-agent model selection. Workload Agent on Anthropic, Test Agent on local Qwen, Reporting Agent on GPT-4o-mini – your choice, agent-by-agent.
Roadmap
Qwen-3.5B for Redmineflux. Specialised LoRA fine-tune. Same training pipeline as our Odient model for Odoo. Release gated on data-quality milestones.
