Image: JetBrains
Why it mattersA CI team that avoided the AI Assistant because it ran on a JetBrains-provided model can now route it through the same provider account it already governs, so the model choice and the billing move under its own controls.
JetBrains released TeamCity 2026.2. Two changes stand out for teams already on the product. Pipelines, the newer way to author CI/CD workflows, has moved from the Early Access Program it entered in 2025.07 to General Availability. The AI Assistant now supports Bring Your Own Key, so an admin can point it at their own Anthropic, OpenAI or Google Gemini account. The post is by Dmitrii Korovin on the JetBrains blog.
Pipelines is now a fully supported option
Pipelines started as a more visual, job-graph-first way to compose CI, aimed at teams that found the older build configuration model too limited. JetBrains says the feature now sits alongside build configurations as a stable option, and this release adds several things the preview was missing.
The most concrete additions are branch handling in Edit mode, a warning when a UI change would try to write to a protected branch, and a Run job even if upstream fails toggle so a job can still execute when its dependencies fail. Promotion, the redeploy-without-rebuild flow, now works for pipelines. Debug lets you re-run a single job inside Edit mode without triggering the whole pipeline. A Without repository option covers pipelines that have no VCS root attached.
BYOK for the AI Assistant
Earlier, the AI Assistant used a model provided through JetBrains. TeamCity 2026.2 adds Bring Your Own Key, and JetBrains lists Anthropic, OpenAI and Google Gemini as supported providers, plus any other supported model. An admin enters the API key in the Admin settings, and requests from the Assistant then flow through that account.
For a team that already has a contract, a data-processing agreement, or a spend budget with one of those providers, the AI Assistant billing and provider choice move under the same controls the organisation uses for the rest of its LLM access. It also unblocks teams that had to leave the Assistant off because the underlying provider was outside the approved list.
New MCP tools, with a manual safety switch
The release adds three MCP tools for pipeline management: teamcity_pipeline_get, teamcity_pipeline_post and teamcity_pipeline_delete. The post and delete tools sit behind what JetBrains calls Brave mode, which an admin has to enable manually before an MCP client can call them. The read tool is available by default.
The split is worth reading carefully. An agent connected to a TeamCity MCP server can inspect pipeline configuration by default, and can change or remove one only after an operator has turned the write path on. The IDE plugin also gains a CLI tool and a simpler MCP server registration in this release.
If TeamCity is already your CI, two things change today. Any pipeline you have been running under the Early Access flag is now on a supported path, so you can pick between build configurations and pipelines on their merits and commit to whichever one fits the team. And the AI Assistant becomes a normal item in your LLM procurement, so you can bring it under the same key-rotation, logging and cost review as your other model use.
The MCP setup is the other thing worth planning around. A read-only agent that can query pipeline state without being able to change it is a useful default, and Brave mode gives an operator a single switch to escalate that when they want it. Any team connecting TeamCity to an agent workflow should treat the write path as an approval step.
Source
Primary source: TeamCity 2026.2: Pipelines General Availability, BYOK for AI Assistant, and More, JetBrains blog, September 2026.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
