Via dailycoin.com
The updated agent framework integrates Slack and Telegram context sharing while maintaining individual identity protections across team environments.
NEAR AI shipped IronClaw 1.2 on August 13, its second major release in under three weeks. The update transforms its open-source Agent OS framework from a solo-user security tool into something more ambitious: a collaborative AI agent layer that lets entire teams share conversations across Slack and Telegram without sacrificing individual security controls.
What IronClaw 1.2 actually does #
The core upgrade centers on how AI agents interact with team communication platforms. IronClaw 1.2 now pulls Slack channel history as context, supporting the last 30 messages or full threads. That means an agent deployed in a team environment can read the room before responding, rather than operating in an information vacuum every time someone pings it.
The previous version required manual allowlists to determine which channels a bot could access. IronClaw 1.2 replaces that friction with presence-based admission, letting bots automatically enter channels and groups based on where team members are active.
Both of these features sit on top of IronClaw’s existing security architecture, which routes every action through memory and approval gates. Each user retains their individual identity within shared conversations, meaning the agent treats permissions and context boundaries as non-negotiable defaults rather than optional settings someone might forget to toggle on.
The framework connects to MCP (Model Context Protocol) tools and apps, providing a standardized way for AI agents to interact with external services while maintaining consistent security guardrails. IronClaw supports deployment either locally or on NEAR AI Cloud, and its tool integrations extend beyond messaging to include email and Notion.
From 1.0 to 1.2 in seventeen days #
IronClaw 1.0 launched on July 27, establishing the foundational guard layer built in Rust that enforces security defaults on every agent action. The architecture prioritized persistent state, meaning agents could maintain context and memory across sessions without losing track of what they’d been doing or who they’d been doing it for.
That first version posted solid benchmark numbers: 93.5% on PinchBench, 88.6% on ClawBench, and 76.4% on OfficeQA.
Both NEAR Foundation and NEAR AI use IronClaw internally for their own security frameworks. The entire codebase remains open source on GitHub, with support across CLI, web, Slack, and Telegram interfaces.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our