Launching today
OpenFang
Open-Source Agent Operating System
238 followers
Open-Source Agent Operating System
238 followers
Open-source Agent OS built in Rust. 7 autonomous Hands that work for you on schedules. 16 security systems. 53 tools. 40 channels. 27 LLM providers. WASM sandbox, Merkle audit trail, taint tracking. Single binary.
RightNow AI
Raycast
RightNow AI
This is the kind of infrastructure play the agent space needs. Most frameworks are still stuck in "prompt-response" mode — basically glorified chat wrappers. But actual autonomy means agents that wake up, do work, and report back without you babysitting.
What caught my attention: the security-first architecture. WASM sandbox + Merkle audit trail + taint tracking isn't just feature creep — it's acknowledging that agents making real-world calls (payments, API writes, data access) need guardrails. At Custyle, we're building AI-powered merchandising agents, and the question of "what happens when an agent goes rogue mid-task?" keeps me up at night.
Quick question: For scheduled agents that fail partway through, is there a built-in retry/recovery mechanism? Or does the Merkle trail just help with debugging after the fact? Also curious how you're thinking about multi-agent coordination — if 7 Hands are running simultaneously, how do they share state without stepping on each other?
JustRecap
hey, congrats on launch. I noticed that you mentioned P2P networking in your website, do you plan to make it distributed in the future?
RightNow AI
@libos Agent frameworks that treat safety as one global toggle break when a Hand has browser access and purchase authority alongside one that just reads APIs. OpenFang splitting guardrails per Hand via HAND.toml is what makes scheduled autonomy actually trustable. Single Rust binary compounds that... no Python dependency tree to audit across 7 autonomous processes on a cron. Where this gets real: overlapping schedules where two Hands compete for the same external resource. That coordination layer is usually where autonomous setups hit their first production stress test.
Really interesting approach building an open-source Agent OS in Rust. The focus on WASM sandbox + security layers stands out. Curious how you’re thinking about long-term extensibility for third-party tools?
The WASM sandbox + Merkle audit trail combo is really smart for agent security. Most frameworks just trust the LLM output blindly - having taint tracking built in from day one shows you actually thought about what happens when agents run unsupervised. Curious how the scheduling system handles failures mid-task.
The security architecture here is impressive - 16 layers including WASM sandboxing, Merkle audit trails, and taint tracking is serious work. Most agent frameworks treat safety as an afterthought, but for agents that actually interact with the real world (payments, database writes, API calls), this is exactly the right priority.
Curious about one thing: when an agent triggers an action with real-world side effects (like sending an email or making a payment), how does the system handle rollback if a later step in the chain fails? Is there a compensation mechanism, or does the sandbox prevent those actions from executing until the full chain is validated?
Congrats on the launch - 4,000+ stars in 4 days speaks for itself.
Congrats on the launch! The idea of building agents that actually run autonomously on schedules instead of just responding to prompts is a really important distinction — most frameworks still feel like glorified chat wrappers. Building this in Rust with a WASM sandbox is a bold choice too. Curious about the security model — with 16 security systems and Merkle audit trails, are you targeting enterprise use cases or is this more for indie devs who want to self-host reliable agents? Either way, shipping a single binary makes the DX so much cleaner.