Moonshot AI has released the latest version of its terminal-first developer tool, Kimi Code CLI, introducing native multi-agent swarm orchestration and enterprise proxy support. The v0.12.0 update transitions several experimental features to default settings, aiming to improve workflow efficiency for developers. Notably, these updates target collaborative intelligence and enterprise-grade deployment.
Orchestrating Agent Swarms in the Terminal
The update introduces the /swarm command, allowing developers to run agent swarms with live progress tracking and rate-limit-aware retries. This capability scales from 100 sub-agents and 1,500 coordinated steps in the K2.5 model to 300 sub-agents and 4,000 coordinated steps in the K2.6 model. Consequently, teams can execute parallel workflows across complex codebases without manual intervention.
In a real-world test, the system completed a 13-hour autonomous rewrite of an open-source financial matching engine. The agent made over 1,000 tool calls across 4,000 lines of code, increasing throughput by 185%. Furthermore, this swarm setup reduces execution time by up to 4.5 times compared to single-agent configurations.
Enterprise Networking with Kimi Code CLI
To support corporate environments, the Kimi Code CLI now honors standard proxy environment variables, including HTTP_PROXY, HTTPS_PROXY, ALL_PROXY, and NO_PROXY. Specifically, the release adds full support for SOCKS5 proxies, which resolves a significant barrier for organizations using legacy or restricted network infrastructure.
This network flexibility is crucial for enterprises with strict security policies. Since the tool runs on Node’s fetch stack, standardizing proxy behavior ensures reliable connectivity behind corporate firewalls. This integration helps maintain strict cybersecurity protocols while utilizing external AI services.
Graduating Experimental Features to Default
Three advanced capabilities have transitioned from experimental opt-ins to enabled-by-default features. These include Goals, Background Questions, and Sub-Skill Discovery. Developers can now use the /goal next command to queue upcoming tasks, which the agent executes automatically upon completing current objectives.
Meanwhile, a new experimental Sub-Skill system has been introduced, allowing users to audit and consolidate existing skills. These updates reflect increased stability in the underlying models. As a result, developers experience less friction during daily coding tasks.
Economic Implications for Regional Tech Teams
For technology teams in Saudi Arabia and the wider Gulf region, deploying the Kimi Code CLI presents a cost-effective alternative to western AI tools. The K2.6 model delivers 80% to 90% of the quality of competing models at approximately 12% of the cost. Specifically, input and output tokens cost $0.60 and $2.50 per million, compared to $5.00 and $25.00 for closed-source alternatives.
This cost efficiency, combined with proxy support, aligns with regional economy goals and digital transformation initiatives. By reducing operational costs, local startups can allocate resources to other development areas.
Source: X (@KimiDevs) — researched
