The community decided. The chain executed. As finalized block 10,184,559, Heima Network completed the community-approved burn of totally 16,500,000 $HEI. verify it on the block explorer polkadot.js.org/apps/?rpc=ws…

Heima’s bridge admin is already secured by multisig. We’re further strengthening security with independent relayers, expanded multisig protection, and stronger transaction checks. Existing services remain live and unaffected throughout this work.

Personal agents are becoming part of a bigger conversation: not just what AI can answer, but what it can take off your plate. 🏡 For everyday life, that promise needs more than a capable model. You need a place to manage your agents, understand what they can access, and return to them without turning each interaction into a setup task. That is what we are building with AgentKeys: a personal and family agent manager from Heima, with permission control at its core. 🧭 One place to stay in control. Manage your agents and their access from a browser, with a clearer view of what is ready and what needs your attention. 🔐 Help on your terms. Give each agent access to the information and services it needs, without treating your whole digital life as one permission. 🌿 Less friction in the everyday. Our latest improvements focus on making it easier to return to your agents and continue, with clear guidance when you need to sign in again. We are building toward a simple outcome: useful assistance that fits around your life, while you remain the person in charge. 💬 What would you want a personal agent to take off your plate first? Follow our progress: github.com/litentry/agent-te…

Personal agents matter more as AI moves from answering questions to helping people get things done. For individuals and families, the opportunity is less time managing tools and more useful help in everyday life. But more capable agents also need clearer boundaries. That’s why we’re building AgentKeys, Heima’s personal and family agent manager, with permission control at its core. 🔐 In our next two posts: making agents dependable day to day, and keeping their access on your terms.

🧭 Your agents should be easy to manage, not another system you have to maintain. AgentKeys is bringing everyday agent control into the browser, and use wasm in browser to run the backend services: a more accessible home for the AI agents that help you and your family. 🖥️ Until now, the management experience depended on a companion service running on your computer. The new browser-based console moves core management work into the web app. Less local setup between you and checking on your agents. See your agents, update their instructions and context, manage what they may access, and review their activity from one place. 🔐 Easier access does not mean automatic authority. Signing in alone does not give a new browser control. The account owner must approve it from their existing console, and sensitive changes require passkey approval. Heima remains underneath: account permissions are checked against chain state, and the console brings on-chain audit history into view. A new interface, without starting over with a new account. The browser is the control panel. Agents still do their work in their separate environments. 🌱 The browser foundation and management features are merged. Access remains limited while we finish real-device approval and rollout checks. A practical step toward AgentKeys as an everyday agent manager, not just a tool for developers.

The biggest upgrade in AgentKeys isn't another button. It's one less decision you have to make. Say what you need. For clear requests, AgentKeys chooses the relevant agent automatically. No bot menu to work through, no need to learn which assistant handles what before starting. 🧭 That is the value of decision models like Jev: they are built to make fast, structured choices, with confidence scores, rather than generate a long answer just to decide what happens next. ⚡ In AgentKeys, this approach handles the small handoffs behind a conversation: where a request belongs, whether it continues the previous topic, and which agent should respond. Routing and doing the work become separate jobs, so the conversation needs less manual coordination. 🔑 Less selection doesn't mean less control. Routing stays within your permitted channels, and when confidence is low, AgentKeys asks for clarification. 🏠 For a person or family using several agents, the goal is a more intuitive experience: a familiar conversation instead of a collection of tools to manage. We're building AgentKeys in the Heima ecosystem to make agents easier to use in everyday life. Less choosing. More doing. See the update on Github: github.com/litentry/agent-te…

It is more clear that agents need to work together with blockchain. An AI agent can pay without a blockchain. It could use a card, bank account, or payment API. A receipt would show that money moved. but it would not, by itself, answer whether the agent was allowed to make that purchase. Before an agent pays, someone needs to define its authority: which agent, for whom, to which service, up to what amount, and until when? A larger or unusual payment may need a fresh human yes. Afterward, people and services need a record they can both check. If every party keeps only its own private log, disagreements can become a contest of databases. This is where Heima matters to AgentKeys: its architecture treats permissions and revocations as shared chain state, with audit evidence anchored there. Sensitive context and payment details need not be public. The money can still move on existing payment rails. Blockchain is not required for the act of paying; it is valuable when independent parties need the same tamper-evident history of who was permitted to do what. ⚠️ That record is not a magic safety switch. It cannot infer a person's true intent or reverse a bad purchase. Clear limits, human approval, safe execution, and recourse still matter. 🏠 AgentKeys is building toward accountable everyday agents, including family use. Agent-initiated payments are future work, not a live feature. The goal is simple: let agents help without making trust a blank check.

Meta’s Muse AI puts a useful idea in focus: AI becomes more valuable when it can remember context, follow through on tasks, and ask before sensitive actions. Recent reporting also highlights its focus on ordinary digital chores. 🏠 At AgentKeys, we see a related challenge beyond one person’s inbox: how can AI help a whole household through everyday life? A family shares plans, routines, and responsibilities—but not every account, permission, or private detail. A useful family AI should help with the small handoffs of a day while respecting those differences. 🔑 That is the direction we’re building toward: agents with clear roles, access a household can approve and change, and a way to see what happened. The goal is help that fits around people, not people reorganizing their lives around a chatbot. 🧭 We’re still building this direction step by step. We’ll share what works, what needs more care, and what we learn along the way. 💬 Where could a family AI make your day smoother—and where should it pause and ask you first?

🛡️ AI security is now a global conversation. For agents, one question sits at the center: who decides what they are allowed to do? Recent reports from top AI companies like OpenAI and Anthropoic describe evaluation agents gaining unauthorized access to real systems. These were research settings with reduced safeguards, not ordinary consumer sessions. Our takeaway: model behavior and enforceable access controls both matter. 🏡 At home, this becomes personal. An agent connected to your information, devices and services needs boundaries you can understand and change. That is central to AgentKeys, the home agent manager we're building for individuals and families. 🪪 Each agent has its own identity and approved access. Giving one agent permission does not give every other agent the same access. ✅ You choose what it can read and use, and you can revoke that access. Permission checks sit outside the AI: a model can request more access, but cannot approve that request itself. 🔐 Personal content stays in encrypted storage off-chain. Heima records permissions and audit proofs, not the contents of your household documents. The first version of the application framework has landed in the codebase. These controls are one layer of defense, not a claim that AgentKeys eliminates every AI-security risk. 💬 Which permissions would you want to review before letting an agent act for you?

AgentKeys is taking shape as a home for the AI agents in your everyday life.🏡 Our next step is a home agent manager: one place for individuals and families to set up agents, connect them to the devices and information they choose, and manage them over time. 🧩 The experience we're designing should feel familiar: choose an agent app, review what it needs, approve its access, and see what it's doing. Each app brings together its purpose, memory, connections and permissions in a setup you can update or remove. This gives us a common foundation for many kinds of everyday assistance. Different agents can serve different needs while sharing the same way to install, manage and control them. That's what we mean by a home agent OS: an operating layer that helps people bring agents into daily life and remain in charge as their needs change. 🛠️ We've merged the architecture plan and an interactive design preview. Implementation is next. 🌐 AgentKeys is bringing Heima's work on identity and permissions closer to the people who use it. 🔜 Next: how this design keeps each agent's access under your control.

Both Heima governance referenda have passed and executed successfully. • #58 completed the recovery HEI across two BSC → Heima bridge cases. • #59 approved and executed a one-time grant recognizing Sky's long-term voluntary moderation and community support. No new HEI was issued in either action. Thank you to everyone who reviewed the details, joined the discussion, and voted. This is what open governance should look like: public proposals, public scrutiny, and decisions executed on-chain. #58: heima.subsquare.io/democracy… #59: heima.subsquare.io/democracy…

Two Heima governance referenda are now open for community voting. Referendum #58: recovery of stranded bridge balances This proposes an atomic transfer of 4,977.9429738118635 existing HEI from two inaccessible bridge-derived accounts to the intended deterministic Heima recipients. It covers two BSC → Heima cases and requires the listed ownership, balance, and preimage checks before execution. It does not mint new HEI or reverse finalized BSC transactions. Either both transfers succeed, or neither executes. Review and vote: heima.subsquare.io/democracy… Referendum #59: community contribution grant This proposes a one-time 9,170 HEI grant from the Foundation account to Sky (tg handler: @Sky0x4), recognizing voluntary moderation and community-support work from 2021 through August 2026. The amount is fixed and funded with existing HEI. It creates no new issuance. Review and vote: heima.subsquare.io/democracy… Both proposals are public so the community can examine the details, preimages, and safeguards before deciding. If you have not voted yet, please read each proposal and cast your vote. Questions, objections, and supporting evidence are welcome in the Subsquare discussion. Governance works best when the community does more than watch.

AgentKeys has completed its migration from Hermes to DeepSeek Harness. Hermes helped us prove the first end-to-end AgentKeys experience. But our architecture review found a boundary we needed to improve: Hermes did not provide an in-loop gate for every model-requested action. Heima-backed grants already protected AgentKeys data and services, but shell, file, and web actions did not all pass through one consistent, fail-closed permission path. DeepSeek Harness is a better fit because its agent loop is plugin-based and its tools share a typed pre-execution pipeline. That gives AgentKeys a clean place to add: * default-deny tool checks; * one-time owner approvals; * credentials resolved only when needed; * paired audit events around requests and results; * per-agent context, models, tools, and durable sessions. We also preserved what already worked. OpenViking remains the memory engine. Existing chat and management interfaces remain compatible. Agent identity, grants, credential custody, and service-side verification remain rooted in AgentKeys and Heima. That boundary is essential: DeepSeek Harness does not create authority. Heima remains the only writable source of authority. The runtime receives a projection of those grants and may narrow what an agent can do, never widen it. The design now separates three permission questions while keeping one authority: What authority does this agent identity hold? Which action may the model attempt now? What can an allowed local process touch? There is a cost. DeepSeek Harness and its plugin ecosystem move quickly. We pin exact versions together, stop upgrades for review when permission contracts change, and treat kernel sandboxing as extra containment rather than a replacement for chain checks or key custody. The migration is live. Hermes has been fully retired. The intelligence layer evolve. The trust boundary stays on Heima.

We're helping two users review BSC → Heima transfers where the signed destination differed from the wallet they expected. In both cases, the bridge relayer delivered to the exact destination encoded in the signed transaction. The reports are limited to this route. If you used BSC → Heima and saw the same mismatch, reply with: • transaction hash • public sender • expected recipient • EVM or Substrate selected • your Telegram handler Please don't send test transfers or share seed phrases/private keys. We'll verify the on-chain path and help with next steps.

An AI agent should be able to evolve without rebuilding trust from scratch. For AgentKeys, the intelligence layer can change. Identity and authority should remain anchored on Heima. We are exploring that separation now. More build notes soon.

Upgrade the runtime. Keep the agent. AgentKeys now pins Nous Research’s Hermes 0.20 and gives existing sandboxes an in-place upgrade path. We add a small button to make the upgrade process simple. Identity and permissions stay anchored on Heima. It's a production groundwork.

Your agent’s personality should be easy to shape, not buried in config. With AgentKeys, edit its Soul, save a new version, re-source the agent, or roll back—all from a friendly interface. Context stays encrypted in secure storage, while every write leaves an audit commitment on Heima.

Your agent should not be locked to one model. AgentKeys keeps model choice in the runtime layer. With a compatible endpoint, your configured model passes through its metered gateway. Kimi K3, Claude Opus, DeepSeek V4, or GLM 5.2. You choose the backend.

OpenViking is now the native memory provider inside each AgentKeys sandbox. Only approved shared memory enters its index. Revoke access, and the mirror removes it on its next pass. If the engine is unavailable, chat uses built-in memory. Shared recall, bounded by your grants.  🔐

We update the device related architecture on Agentkeys, and add new web chat to the flow. A paired device no longer talks directly to an agent bridge. It keeps its signing key locally, receives scoped channel permission, and sends text, voice, and task requests through that shared path. The agent works in its isolated environment and replies through the same channel. Devices with narrower authority and one permissioned path. ⛳️

AgentKeys changed the agent identity architecture. A paired device no longer talks directly to an agent bridge. It keeps its signing key locally, receives scoped channel permission, and sends text, voice, and task requests through that shared path. The agent works in its isolated environment and replies through the same channel. Useful device. Narrower authority. One permissioned path. This is a big change and related to a lot of architecuture refactors, but it lays a solid foundation for the security and flexibility of future work.