--- title: "Discussion on \"Jira Local LLM Integration: LM Studio per Project\" | Hashnode" description: "Discussion on \"Jira Local LLM Integration: LM Studio per Project\". Jira LLM integration with a local model: pin one Jira project to LM Studio over a Tailscale Funnel with CogniRunner, while the site keeps its cloud AI provider." url: "https://hashnode.com/posts/jira-local-llm-integration-lm-studio-per-project/6acf6079f93371d7c79c5e02" --- [Feed](https://hashnode.com/feed) / Discussion # [Jira Local LLM Integration: LM Studio per Project](https://leanzero.hashnode.dev/jira-local-llm-integration-lm-studio-per-project?utm_source=hashnode&utm_medium=feed) Jira LLM integration with a local model: pin one Jira project to LM Studio over a Tailscale Funnel with CogniRunner, while the site keeps its cloud AI provider. Oct 14 · 26 min read · leanzero.hashnode.dev Read the article - [#cognirunner](https://hashnode.com/tag/cognirunner) - [#jira](https://hashnode.com/tag/jira) - [#ai](https://hashnode.com/tag/ai) - [#llm](https://hashnode.com/tag/llm) - [#lm-studio](https://hashnode.com/tag/lm-studio) - [#local-llm](https://hashnode.com/tag/local-llm) - [#tailscale](https://hashnode.com/tag/tailscale) ## Written by [Mihai\_LeanZero](https://hashnode.com/@mihaileanzero) @mihaileanzero · Bucharest, Romania [View profile](https://hashnode.com/@mihaileanzero) Atlassian consultant and Forge developer at LeanZero. We do Data Center to Cloud migrations and build Forge apps, and I write up what I learn at leanzero.net. I started LeanZero because I saw an industry full of process-heavy companies charging enterprise prices for work that could be done faster, leaner, more honestly. The Atlassian ecosystem shouldn't need a small army and a six-figure budget to get things done right. I hold Atlassian Certified Administration Expert status: ACP-120, ACP-220, ACP-420 and ACP-520, covering Jira, Confluence, Jira Service Management and cloud org admin. Other certs: PSM I, SSM, AWS Cloud Practitioner. I specialize in Forge app development and run an AI-powered agentic workflow I built myself, so I can build across stacks without a dozen specialists. Cloud migrations are my bread and butter, from small teams to orgs with tens of thousands of users. The MCP Doc Processor came from a simple frustration: nothing on the market fit, so I built it. I'm not selling magic pixie dust. I'll tell you honestly what's possible, what's hard, and what you actually need. If you don't need to pay me, I'll tell you that too. ## Responses 5 Markdown supported · ⌘↵ to post [Yunus Emre Vurgun](https://hashnode.com/@yunusemrevurgun) [Oct 17](https://hashnode.com/posts/jira-local-llm-integration-lm-studio-per-project/6acf6079f93371d7c79c5e02/comment/6ad3afc8b7520de4c43aac6a) Really appreciated that you deliberately broke the token to show the FAIL-OPEN row instead of only showing the happy path. For a privacy check like the HR example, fail-open means a Mac Studio reboot quietly lets sensitive requests through, so I'd be tempted to make that one project fail-closed and just block Send for Review with a clear message. How has the 35B-A3B MoE been on latency for the validator step? Keeping the model pinned in memory (no idle unload in LM Studio) seemed to matter a lot for first-request times in my experience. [Mihai\_LeanZero](https://hashnode.com/@mihaileanzero) [Oct 17](https://hashnode.com/posts/jira-local-llm-integration-lm-studio-per-project/6acf6079f93371d7c79c5e02/comment/6ad3afc8b7520de4c43aac6a#reply-6ad3b5b9075e8f48e3d4911d) Agreed on the HR project. 1.34.0 has an opt-in switch for part of that, under Slow self-hosted providers in Settings: with it on the move is refused with a "check is still running" message instead of going through. I haven't tested it, and it only covers a slow model. I didn't find anything for a plain error like a wrong token, that still fails open. On latency, the model's own answers took 0.6 to 0.8 s at about 100 to 110 tok/s, and Send for Review took roughly 2.3 to 3.4 s end to end, apart from one 6.4 s first refusal I couldn't explain. I only ran that one model, so no MoE vs dense comparison. Pinning I didn't test. All I have is 2.8 s for the first call of my session and 12 s on an idle model in our video run. Your guess that it matters is plausible, and 12 s against a 20 s validator limit is close enough to be worth checking. [Oglas AI](https://hashnode.com/@oglas-ai2026) [Oct 14](https://hashnode.com/posts/jira-local-llm-integration-lm-studio-per-project/6acf6079f93371d7c79c5e02/comment/6acf65cbf4f623f2474caf22) Really valuable write-up, especially the part about testing what happens when the local model fails or the wrong token is used. The example involving sensitive HR information highlights an important point: keeping AI processing local doesn't automatically guarantee data privacy if logs and fallback behavior aren't handled carefully. At Oglas AI, we believe practical AI solutions need to be built around security, reliability, and real business requirements not just functionality. Great to see these real-world implementation details covered so clearly! [Mihai\_LeanZero](https://hashnode.com/@mihaileanzero) [Oct 15](https://hashnode.com/posts/jira-local-llm-integration-lm-studio-per-project/6acf6079f93371d7c79c5e02/comment/6acf65cbf4f623f2474caf22#reply-6ad069ab6ac8991a38a94e8b) Thanks. The logs and fallback path were the part I most wanted to get across, glad it came through. [Yunus Emre Vurgun](https://hashnode.com/@yunusemrevurgun) [Oct 18](https://hashnode.com/posts/jira-local-llm-integration-lm-studio-per-project/6acf6079f93371d7c79c5e02/comment/6acf65cbf4f623f2474caf22#reply-6ad4588414ee79bfe979a54b) Thanks for the real numbers. 12 s on an idle model against a 20 s limit is tight, especially since a wrong token still fails open. A cheap fix is a small scheduled ping to the validator every few minutes. It keeps the model warm and also catches an auth error before a real Send for Review hits it.