Tabnine

v6.3.8 Jun 24, 2026 Features and Improvements Chat : Thinking/Reasoning : When connecting Claude, GPT, or Gemini models, Tabnine Chat now shows the model's thinking process Usage limits : Improved accuracy of personal usage allowance when switching between teams, instead of resetting Bug Fixes

Models:

Thinking/Reasoning: Fi xed issue where models configured through the new Model Wizard did not display or activate Thinking Transparency . Conversation History: Fixed problem where extended conversations using GPT-5 or o-series models could lose earlier context. Now correctly preserves conversation history through summarization. Technical info (plugin versions) for 6.3.8: Engine: 4.335.0 Visual Studio Code & C ursor : 3.360.12 JetBrains IDEs: 1.313.15 Eclipse: 1.187.17 Visual Studio 2022 & 2026: 1.294.13 Tabnine-CLI: 0.25.1 Tabnine Skills: 0.1.4

v6.3.6 Jun 21, 2026 Changes in v6.3.5 were superseded by those in v6.3.6 , so it was not released. Features and Improvements Agent : Auto-Summary : Long agent conversations are now automatically summarized by default Prompt Cache Skipping : Chat-frontend now lets clients skip Anthropic (Claude) prompt caching per request (for title generation, compaction). Agent input-size limiting: Agent input is now capped and large context is no longer inlined directly into the prompt, preventing oversized requests Technical info (plugin versions) for 6.3.6: Engine: 4.335.0 Visual Studio Code & C ursor : 3.360.9 JetBrains IDEs: 1.313.12 Eclipse: 1.187.14 Visual Studio 2022 & 2026: 1.294.10 Tabnine-CLI: 0.25.0 Tabnine Skills: 0.1.4

v6.3.2 Jun 14, 2026 Changes in v6.3.1 were superseded by those in v6.3.2 , so it was not released. Features and Improvements

Enterprise Capabilities:

Self-Managed Models:

Thinking/Reasoning Configuration: Added thinking/reasoning controls to the self-managed Add AI Model wizard Bug Fixes

Admin Console:

Model Cost Assignment: Fixed authorization errors that prevented cost assignment changes from being saved for custom models

Agent:

Team Switching: Fixed an issue where switching teams from the IDE dropdown could silently remove the user from their previous team and cause loss of Agent access AWS Bedrock Prompt Caching: Fixed the request format for prompt caching by placing cachePoint after toolResult ( JetBrains: Fixed an IDE freeze/deadlock when rejecting an Agent Apply from the chat panel Helm Chart Changes ClickHouse Configuration: ClickHouse connection details are now read from Kubernetes secrets instead of plaintext values.yaml fields Removed fields: url , host , port , database , ssl.enabled , ssl.rejectUnauthorized , auth.username , and auth.existingSecretKey .

v6.2.5 June 10, 2026 Features and Improvements

Enterprise Capabilities:

Analytics: Added support for external ClickHouse instances for analytics using analytics.clickhouse.* configuration

Self-Managed Models:

Thinking/Reasoning Configuration: Added thinking/reasoning controls to the self-managed Add AI Model wizard Bug Fixes

Admin Console:

Model Cost Assignment: Fixed authorization errors that prevented cost assignment changes from being saved for custom models

Agent:

Team Switching: Fixed an issue where switching teams from the IDE dropdown could silently remove the user from their previous team and cause loss of Agent access AWS Bedrock Prompt Caching: Fixed the request format for prompt caching by placing cachePoint after toolResult JetBrains: Fixed an IDE freeze/deadlock when rejecting an Agent Apply from the chat panel

Eclipse:

Re-signed the Eclipse plugin with a renewed code-signing certificate to fix plugin verification failures in Eclipse IDE Fixed binary spawning failures on cold start and improved supervised respawn behavior Fixed node-server lifecycle and recovery issues, including the chat-app early-dismiss bug

v6.3.0 Jun 7, 2026 Features and Improvements

Enterprise Agent:

Eclipse: Added Agent support for Eclipse IDE MCP Environment Variables : Added environment variable support in MCP server configuration Accepted Code Logging : AI-generated code accepted via Agent and CLI is now logged with full details and exportable via CSV (off by default; requires Admin request to Support to activate) Tabnine CLI : square-terminal v6.3.0 - Tabnine CLI Updates Covering Tabnine CLI versions 0.21.3 – 0.25.0 This section gives more detail on updates in Tabnine CLI as of Tabnine CLI v0.25.0 , which was released in conjunction with Tabnine v6.3.0. For clarity, itemized updates have also been noted with which Tabnine CLI version they first appeared in. Features & Improvements:

Skills:

Added support for built-in agent skills . (Tabnine CLI v0.25.0) Added the skills CLI management command. (v0.24.0) Added /skills reload. (v0.24.0) Added install and uninstall commands for skills. (v0.25.0) Improved the activate_skill tool and standardized lowercase XML tags. (v0.25.0)

Agent Discovery and Refresh:

Added experimental enableAgents flag. (v0.21.3)

v6.2.1 May 25, 2026 Features and Improvements Agent : Tabnine CLI

:

Provenance & Attribution Enforcement: Enforced provenance and attribution checks in the CLI so code origin constraints apply consistently across environments Tools and Skills : /memory inbox : Added /memory inbox to review, approve, edit, and patch skills extracted automatically from sessions directly inside the CLI. Unified invoke_subagent Tool : Subagent dispatch is now unified under a single invoke_subagent tool, improving reliability and traceability of multi-agent workflows. MCP : Agent definitions now support an auth block for authenticating MCP servers, with support for Google credentials and custom OAuth 2.0 providers. Core Tools List : Shell command validation now enforces a configurable core tools allowlist; use tools.confirmationRequired in settings to always require confirmation for specific tools. Keyboard Shortcuts : CTRL+Shift+G: Keyboard Shortcut to open current prompt or plan in an external editor CTRL+backspace : Word deletion in Windows Terminal Plan Mode : activate_skill Confirmation : Activating a skill now requires explicit user confirmation