RightNow-AI/openfang

every new "agent framework" is just another python wrapper around langchain pretending to be an OS. openfang is actually written in rust from scratch. works offline, runs on a pi if you want! not a framework. it's the best OS for agents github.com/RightNow-AI/openf…

Matched through a repo-specific project phrase query.

OpenFang v0.5.10 is out! here's everything since v0.5.7: - AWS Bedrock provider with native Converse API, bearer auth, and streaming - GitHub Copilot OAuth device flow rewrite with token persistence and refresh rotation - Novita AI provider - Russian localization, complete i18n framework with English and Russian out of the box - multi-instance hands, run multiple instances of the same hand type with optional instance_name - SearXNG improvements, simplified config, new list_searxng_categories tool - WeCom Stream mode with WebSocket long-connection, protobuf framing, and heartbeat - Gemini embedding driver with native batch embedding and automatic fallback - session interaction endpoints for per-user conversation isolation in multi-tenant setups - armv7 target for Raspberry Pi and embedded ARM boards - Revolt adapter now supports self-hosted instances with configurable ws_url and allowed_channels - auth fail-closed, non-loopback requests rejected with 401 when no api_key configured, set OPENFANG_ALLOW_NO_AUTH=1 for LAN access - config persistence finally works, dashboard changes to agent model, provider, fallback chain, system prompt write back to agent.toml on disk - agent context.md re-read on every turn so cron-updated data is picked up immediately - multimodal user messages now combine text and image blocks into a single message, vision models no longer receive images with no prompt - schedule tools and HTTP routes now actually fire, schedule_create/list/delete go through the kernel cron scheduler with legacy migration - per-channel agent name prefix on outbound messages, set prefix_agent_name = "bracket" or "bold_bracket" for multi-agent channels - openfang hand config <id> command wired up with --get, --set, --unset, --list - openfang config get default_model.base_url works again with 8 new unit tests - [SECURITY] rm bypass in Allowlist mode closed, process_start now validates command and args - [SECURITY] PowerShell command bypass closed, inner commands extracted and validated against exec allowlist - [SECURITY] dashboard auth upgraded to Argon2id with random salts - fix global [[fallback_providers]] actually used at runtime, network errors escalate to fallback - fix Discord gateway heartbeat with ACK gate and zombie connection detection - fix system prompt no longer leaks in Web UI session endpoint - fix custom hands persist across daemon restart - fix WebSocket auth with special characters in API keys - fix Moonshot session repair, deduplicate_tool_results runs before insert_synthetic_results - fix Moonshot Kimi K2.5 routing to correct API domain - fix Nextcloud chat polling endpoint - fix Qwen/OpenAI-compat tool_calls orphaning after context overflow - fix LINE webhook signature validation - fix exec_policy mode=full now bypasses approval gate - fix install script fetches latest release version correctly - fix session endpoint returns full tool results (removed 2000-char truncation) - fix Gemini streaming SSE parser handles \r\n line endings - fix tool_use.input always normalized to JSON object - fix heartbeat skips idle agents, no more crash-recover loops - fix hand continuous interval changed from 60s to 3600s - fix chat streaming renders in real-time - fix cron job preservation across Hand reactivation - Signal defaults to plain text - new agents use configured provider instead of hardcoded Groq - intermediate tool text no longer dropped - 45 bugs fixed, 86 issues closed, 85 PRs merged, zero open bugs remaining github.com/RightNow-AI/openf…

Matched through a repo-specific project phrase query.

things you can do with OpenFang: make an agent that generates your leads while you sleep make an agent that manages your Twitter while you touch grass make an agent that turns your 2hr podcast into 30 viral clips make an agent that monitors your competitors so you can pretend you don't care It's open source and written in Rust github.com/RightNow-AI/openf…

Matched through a repo-specific project phrase query.

Maybe a lot of you are feeling unsure or kinda scared about $openfang , but here’s why I still believe in it: -fees totaling 131 SOL haven’t even been claimed yet -the dev is always active and keeps building - @openfangg has more advantages compared to other projects -the community is solid We never really know what’s gonna happen with a project or its coin in the future, but honestly, with the current market cap, I don’t see any reason not to buy or even be afraid to hold it. Better late than never. @Akashi203 openfang.sh/

Matched through a repo-specific project phrase query.

My Hermes agent Jimmy impresses me more and more every day. I have been running openfang since it launched dropped openclaw early was just chewing through tokens. You have done a great job. I am going to be migrating my whole fleet over shortly!

Matched through a repo-specific project phrase query.

Still deciding which framework? OpenClaw: "I want my AI to be a reliable, always-available team member across every channel" Agent Zero: "I want my AI to write code, browse the web, and automate complex workflows autonomously" OpenFang: "I want an army of specialist agents working around the clock" Hermes: "I want an AI that learns me — my style, my patterns, my preferences" All 4 included. Switch anytime. Your data follows you. aibar.ai/waitlist

Matched through a repo-specific project phrase query.

No CCP ties to OpenFang. Deep search (web, GitHub, X, LinkedIn, company filings): - Solo-built by Jaber J. (Akashi203), Jordanian founder of RightNow AI (Delaware-registered US entity, NVIDIA Inception member). - GitHub RightNow-AI/openfang: 1 contributor (jaberjaber23), MIT license, 137k LOC Rust from scratch, launched Feb 2026. - No Chinese devs, funding, Beijing/Shanghai links, or code origins. (Unrelated old openfang camera project ignored.) Supports some CN LLMs/channels (Qwen, Zhipu, DingTalk) for broad compat—like every major agent tool. Fully auditable repo. Clean.

Contains the exact GitHub repo slug.

We’re underestimating what changes when agents run 24/7. No human in the loop = new threat model. OpenFang gets this: → Actions require approval gates → Financial ops are restricted by default → Write permissions aren’t assumed That’s not cautious design. That’s the only way autonomy works in the real world.

Matched through a repo-specific project phrase query.

Most agent frameworks optimize for demos. OpenFang optimizes for survival in production. That’s the real shift. → WASM sandboxing → Audit trails (Merkle chains) → Prompt injection scanning at runtime This isn’t “extra security.” It’s security as architecture. Because autonomous agents don’t fail loudly. They fail silently at scale.

Matched through a repo-specific project phrase query.