Hermes Agent v0.14.0: The Foundation Release — 19-Second Cold Start Cut, 22 Platforms, 12 Security Fixes
NousResearch has shipped Hermes Agent v0.14.0, titled "The Foundation Release," just nine days after v0.13.0. The pace signals active development, and the changelog justifies the name: this release addresses performance, connectivity, and security simultaneously in ways that make v0.14.0 the more appropriate starting point for new Hermes deployments.
The headline performance gain is a 19-second reduction in cold-start time — a meaningful improvement for interactive use cases where the agent is initialised frequently. Browser console performance jumps 180x, which matters for the growing share of Hermes users driving it through web-based interfaces. The update also adds cross-session Claude prompt caching, reducing both latency and token costs for repeated interactions with similar context. LSP-based semantic diagnostics bring code-aware error detection to the tool invocation layer, enabling Hermes to reason about code structure rather than treating it as opaque text.
Connectivity reaches 22 platforms in v0.14.0, adding Microsoft Teams, LINE, and SimpleX Chat to an already broad roster. Native button UI support arrives for Telegram and Discord channels, improving the interaction model for agents deployed in those environments. The /handoff command enables mid-conversation model switching — developers can start a session on one model and transfer context to another (including the newly added xAI Grok with its 1M-token context window) without restarting. An OpenAI-compatible local proxy ships alongside the model additions, allowing third-party tools that speak the OpenAI API format to connect to Hermes without custom integration work.
The 12 P0 security fixes make v0.14.0 a mandatory upgrade for any production Hermes deployment. The most operationally significant is protection against sudo brute-force attacks, relevant for agents that are granted elevated permissions on developer machines. Combined with the eight security closures shipped in v0.13.0, the framework has addressed 20 security issues in two weeks — a pattern worth tracking as Hermes gains adoption in more sensitive environments.
Read more — NousResearch
xAI Launches Grok Build: Agentic CLI Coding Tool Enters Beta
xAI has released Grok Build, a command-line agentic coding tool that enters a field already occupied by Claude Code, OpenAI Codex, and Gemini Code Assist. The timing is notable — xAI joins the agentic coding race weeks after its merger with SpaceX, with Grok Build serving as the new combined entity's first dedicated developer product.
Grok Build operates as a CLI agent capable of multi-step planning, automated file modification, and parallel subagent delegation. The parallel delegation feature is the most technically distinctive: when a task can be decomposed into independent subtasks — implementing two separate features, running tests while scaffolding code, generating documentation alongside implementation — Grok Build can spin up multiple subagent instances working concurrently, then merge results. This positions it closer to multi-agent frameworks than single-turn coding assistants.
The initial beta is available exclusively to SuperGrok Heavy subscribers at $300 per month, making it the most expensive tier among the current agentic coding tools at launch. xAI has not published a roadmap for broader availability or a lower pricing tier. The $300 price point and SuperGrok Heavy gating suggest the beta audience is primarily developers who are already deep in the xAI/Grok ecosystem and willing to pay premium prices for early access. For teams evaluating their agentic coding tool stack, Grok Build is worth watching but not yet ready for broad enterprise evaluation — the combination of beta status, limited availability, and opaque roadmap makes it premature to use as a primary tool.
Read more — CIO Dive
Links & Sources
You May Also Enjoy
