Ai

CowAgent 2.0.7 Adds Image Generation Skills and Broader Model Support

CowAgent 2.0.7 Adds Image Generation Skills and Broader Model Support

CowAgent 2.0.7 is a meaningful AI platform update centered on multimodal creation, model flexibility, and usability improvements. The release adds a built-in image generation skill with automatic provider routing, expands support for new large models, improves knowledge base structure handling, refines dream-memory organization, and fixes several web console issues that affect day-to-day operator experience.

What Changed

The headline feature in version 2.0.7 is a new built-in image generation skill. CowAgent now supports text-to-image, image-to-image, and multi-image fusion workflows through a unified capability that can automatically route across six providers. The routing chain includes OpenAI GPT-Image-2, Gemini Nano Banana, Seedream, Qwen, MiniMax, and LinkAI, allowing deployments to fall back across vendors without manual model switching.

This image workflow is designed to be usable out of the box once API keys are configured. Users can still explicitly select a model in conversation, but the default experience now favors automatic routing. The skill also exposes practical control parameters such as quality, resolution, and aspect ratio, with vendor-specific compatibility handled behind the scenes. In addition, passing existing images enables editing, style transfer, and multi-image composition, which broadens CowAgent from pure chat orchestration into more complete creative automation.

Version 2.0.7 also expands the model roster. The release adds support for Kimi K2.6, Claude Opus 4.7, and GLM 5.1, while also introducing Kimi Coding Plan mode. Another notable addition is support for custom model providers, which should make it easier for teams to connect local model infrastructure or integrate less common vendors without waiting for first-party presets.

On the knowledge side, CowAgent improves how its knowledge base handles nested directories. Multi-level folder structures are now supported in listing and display views, and root-level files such as index.md and log.md are shown in the knowledge tree. The release also fixes empty-state statistics so root-level files no longer distort whether a knowledge base appears populated.

The dream-memory subsystem was refined as well. Memory files are now automatically organized by date, creating a clearer archival structure, and scheduled triggers gain random jitter to reduce collision risk in clustered environments.

The skill system received several quality-of-life improvements. Running the /skill command now refreshes and reloads the latest skills automatically, and skill installation supports a wider range of source formats, including URLs, zip archives, and local files.

Finally, the web console sees a broad set of interface fixes. Chat auto-scroll behavior was adjusted so users reviewing earlier content are not forced back to the bottom, mobile sidebar handling was improved, duplicate image rendering in a single message was fixed, long reasoning output is truncated to reduce frontend lag, and session title fallback logic was repaired. The release also fixes a Gemini tool-call issue where results were not being returned and preserves tool_calls during agent retries after empty responses.

Why It Matters

This release matters because it strengthens CowAgent in three practical areas at once: multimodal capability, deployment flexibility, and operational polish. The new image generation skill is the clearest strategic step, giving teams a built-in path to visual workflows without requiring separate tooling or brittle manual provider selection.

The broader model support also makes CowAgent more adaptable for organizations that want optionality across AI vendors, including local or custom backends. That is especially relevant for enterprise and developer teams trying to balance cost, availability, and model specialization.

Meanwhile, the knowledge base and dream-memory changes improve information hygiene rather than just adding features. Better nested-directory support and cleaner date-based memory organization should reduce friction for teams running larger or longer-lived agent environments.

Overall, CowAgent 2.0.7 is less about a single isolated fix and more about making the platform more capable and more production-friendly. The release pushes CowAgent further toward being a flexible, multi-provider AI agent framework that can handle both conversational and creative workflows with fewer manual workarounds.

Official Source: https://github.com/zhayujie/CowAgent/releases/tag/2.0.7