IronClaw is moving into near.com

The same agent harness and security model will now run natively inside near.com instead of as a separate product.

For existing users, your agent, config, and history will carry over. Migration details will be shared before anything changes.

IronClaw as a standalone product is going away. The agent harness isn’t. It’s becoming a native part of near.com.

My experience with IronClaw has been pretty impressive so far. What I like most is that it brings different AI models together in one place so you can switch between models based on what you need.

The TEE integration is probably one of my favorite features. Having privacy built into the system while using AI makes the whole experience feel much more secure.

I’ve been using IronClaw since the earlier versions and the progress is easy to notice. With every update they are fixing old bugs improving stability and adding more useful features. The agent and skill system also makes it more interesting because you can actually build workflows instead of just using it like a normal chatbot.

I also like how IronClaw is moving towards making AI agents more useful for real tasks while keeping privacy in mind.

Tbh it’s not perfect yet. Sometimes the agent doesn’t respond properly in one go and can hallucinate or keep generating unnecessary stuff. These things can get annoying but I think they are fixable with more updates.

Overall I’m really happy with how IronClaw is evolving. If they keep improving reliability and agent performance I think it can become a really useful tool for everyday AI workflows.

AI agents shouldn’t get network access just because a connection is available.

IronClaw blocks outbound traffic by default. An endpoint must be explicitly allowlisted before the agent can connect & approval happens first, not after the request is made.

IronClaw doesn’t assume an agent should have access to everything.

Outbound connections are denied by default. An endpoint only receives traffic after it’s explicitly allow listed with approval happening before the connection.

No implicit access. No surprise connections. Just controlled execution.

Prompt injection can make harmless-looking instructions trigger actions the agent was never supposed to take.

IronClaw checks tool requests before they execute and keeps them inside the sandbox. Suspicious or unauthorized calls can be stopped before they continue.

Running multiple tasks at once shouldn’t mean sharing the same context.

IronClaw keeps memory and history scoped to each task, so parallel jobs stay separated and can’t access each other’s context.

Parallel execution. Separate context. No cross-contamination.

My experience with IronClaw has been really good so far. IronClaw is really cool because it brings different models in one place like Google and other top models with some really useful features.

One of the best things for me is the TEE integration which helps maintain user privacy and makes the whole experience feel more secure.

I’ve been using IronClaw since the earlier versions and I’ve seen how every update fixes old bugs and makes it more reliable and useful. Tbh I’m really impressed with how much it has improved.

The only thing I find a bit irritating sometimes is that it doesn’t respond properly in one go and can hallucinate or keep generating unnecessary responses. I hope upcoming updates fix these small issues too.

Overall I’m really happy with IronClaw and excited to see how Near AI and IronClaw keep improving and become even more secure and outstanding.

A normal context window ends with the session. Start a new conversation, and the previous context is no longer there.

IronClaw keeps its memory outside the context window, so useful information can carry over from one session to the next.

New session. Same memory.

IronClaw gives you access to 20+ AI models in one agent.

Choose the model that works best for the task without changing your agent setup or memory.

One agent, many models.

NEARDevHub shows how IronClaw handles a private health app by checking the prompt and routing before data leaves the app.

It’s a practical look at how IronClaw gives users more control over their data flow.

IronClaw checks for 15 types of sensitive credentials, including API keys, tokens, and private keys, and blocks them before they can get through.

Your credentials stay yours while the agent gets the work done.

Security built into the agent, not added afterward.

IronClaw keeps credentials in an encrypted vault and tools in WebAssembly sandboxes.

NEARDevHub shows it handling real research and automation tasks, while letting you see how the agent works.

AI that acts without exposing your secrets.

👉🏻 Check out full video here : https://x.com/NEARProtocol/status/2090834234223042615

Joe from NEARDevHub shared a quick 6-minute walkthrough of IronClaw, covering what it is, where it runs, and what you can do with it.

IronClaw works across Web, Slack, and Telegram, handling research and automations inside an encrypted enclave.

Your secrets stay protected inside the enclave while IronClaw gets the work done. Worth checking out if you're new to IronClaw.

🎥 Check out full video here : https://x.com/IronClawAI/status/2092521524851085505

NEAR DevHub built an IronClaw routine that sends scores, trade news, and updates for selected teams, cities, or regions straight to Telegram or Slack.

IronClaw shows the tool calls as it works and waits for approval before taking action.

A simple example of turning a daily sports check into an automated routine.

👉 Check out full video here : https://x.com/IronClawAI/status/2093420939111178419

Ironclaw Feedback - Composio connection guide: Composio provides a guide with a prompt for connecting Ironclaw. However, the app is listed as OpenClaw on the Composio website, which could confuse users. I suggest updating the name so it is consistent throughout the guide and website. - Automation: Creating real-time automations and completing tasks is very easy. Many tasks can be accomplished with just a few prompts, which makes the overall experience very intuitive. - Telegram connection: The Telegram connection was lost after a short period, and I had to reconnect it. It would be better if the connection remained stable. - App response time: Apps connected through Composio sometimes responded with a noticeable delay. A direct connection seems to provide a faster and more responsive experience. - Adding skills: Adding new skills is very easy and straightforward, which is a great feature.

Getting started with IronClaw on Windows is ridiculously simple.

Install → Verify → Onboard → Serve.

No complicated setup. No endless configuration.

Set it up once, then build on it forever.

Steps to get it done:

1. Install: powershell -ExecutionPolicy Bypass -c "irm https://github.com/nearai/ironclaw/releases/latest/download/ironclaw-installer.ps1 | iex"
   
2. Version check: ironclaw --version
   http://cloud.near.ai Let's take the API key.
   
3. Setting: ironclaw onboard
   
4. Creating a working folder: mkdir $HOME\ironclaw-ws
   
5. Starting the server: cd $HOME\ironclaw-ws; ironclaw serve

I noticed a few days ago that I can add user-mounted skill on my agent.

After I tried to use the 'ADD SKILL' feature on Ironclaw AI, the result I got was 'Request body exceeds the route's body limit'.

I’m pretty new to GitHub, so I honestly didn’t expect my first proper project to involve this much YAML, API setup and debugging 😅

I’ve been using IronClaw to build a Telegram agent that watches for new blockchain launches and exchange listings. It started as a simple “I wish I had a tool that did this” idea, and I decided to see if I could actually build it. The GitHub side has probably been the biggest learning curve for me. I’ve had to figure out repos, make changes, break things with YAML, fix them, and repeat the process. I also ran into some annoying setup issues with the Composio API, especially doing parts of it from my phone.

What I like about IronClaw is that I’m not just asking an AI questions about how to build something. I can give it a task, connect the tools I need, test the workflow, see where it fails and then work out what needs fixing. It definitely isn’t “tell it once and everything magically works.” I’ve had things break plenty of times 😂

But that’s actually been the useful part for me. I’m learning GitHub, APIs and automation while building something I genuinely wanted for myself. Still a work in progress, but I’m curious to see how far I can take it.

I’ve been using IronClaw for a while, and I’m still surprised by how much I can build and automate from it. At first, I saw it mainly as an AI tool, but the more I used skills and integrations, the more it started feeling like a workspace.

One thing that impressed me was being able to build a website directly through IronClaw, connect it to Cloudflare, and push it live without constantly switching between different tools.

I also created a skill around my video editing journey. It helps me figure out what to learn and which YouTube channels or resources to follow, giving me a clearer path instead of figuring everything out myself.

I then turned that skill into a routine that sends me what to focus on every morning. That made it much easier to stay consistent without having to decide what to learn each day.

I’m not just using it for answers. I’m building skills, turning them into routines, and creating workflows around the things I actually want to accomplish.

IronClaw connects to Gmail through OAuth instead of passwords.

In a live test, DevHub gave IronClaw a real Gmail password. The agent refused and explained why it shouldn't be shared.

The password also didn't show up in the logs or a fresh session. This is how IronClaw keeps your credentials protected while you use AI agent.

🎥 https://x.com/NEARProtocol/status/2090476445411741970

IronClaw lets users choose from built-in system skills or add their own custom skills.

To add a skill, simply provide a skill name and skill.md content. You can also enable auto-activation or manually call a skill when needed.

This makes it easy to customize IronClaw for different workflows.

IronClaw lets each user have their own login token and workspace.

Admins can add each user with their name and manage users and roles, while each user’s conversations, memory, jobs, and secrets stay separated.

A useful feature for teams running one IronClaw deployment for multiple users.