[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f35lka0nvaky63":3},{"code":4,"message":5,"data":6,"timestamp":39},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":22,"summary_md":23,"topic":24,"github_topics":27,"license":35,"homepage":36,"updated_at":37,"history":38},"TencentCloud\u002FOctop",7,"A smarter, self-hosted AI assistant — multi-user, multi-agent.","Python",10916,1407,6225,108,[16],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},49976872,"Octop, a self-hosted AI assistant by Tencent",1,3,"2026-10-06","A self-hosted multi-user, multi-agent assistant platform for households and small teams that want one private instance.","## The problem\n\nMost AI assistants assume one person, one chat window, and a cloud vendor holding the history. Teams and families that want several people sharing agents, channels and long-lived memory end up stitching together a chat UI, an orchestration layer and a pile of message-bot glue. Octop ships all of that as one self-hosted process: a dashboard, CLI, IM bridges, cron and a per-agent workspace, with state in a local database under `~\u002F.octop\u002F`.\n\n## What's different\n\nThe usual route is a hosted assistant client, or a framework like LangChain plus your own glue. Octop instead ships an opinionated stack (Harness, Gateway, Memory, Browser as separate repos) composed into a single restart-safe process with no external queue or broker — every surface (Web, IM, cron) goes through one in-process processor. It also treats agents as long-lived *experts* with their own workspace, model providers, channels and cron jobs, rather than as sessions in a shared pool. The other distinctive choice is breadth of integration: eight IM platforms plus Feishu\u002FDingTalk\u002FWeCom, OAuth and MCP connectors, and bidirectional ACP so the same agent can be used from an IDE or delegated to Claude Code \u002F OpenCode with permission gates.\n\n## Who it's for\n\nSmall teams and households that already run a server or NAS and want one admin account managing several users' agents, with data kept local. It fits people who want WeChat\u002FFeishu-bound assistants and scheduled pushes without writing the bot layer. Developers looking for a library to embed in an existing product should look elsewhere — this is an application, not a composable SDK, and its opinionated single-process design is a constraint, not a feature. Those wanting a polished cloud assistant with no ops work are also in the wrong place.\n\n## Cost of adoption\n\nPython 3.12+, FastAPI\u002Fuvicorn, and a first-run `octop init` wizard; MIT licensed. The real cost is operational: you host it, you own the model provider keys, and you handle upgrades. The README is broad but shallow on specifics — the security claims (JWT isolation, shell guardrails, PII redaction) have no detail on how they are enforced, and several headline features (AgentTeams, cloud collaboration) are marked beta or sketched in one line. It depends on four sibling repos from the same org, so the supply chain is narrow and version-locked.\n\n## Verdict\n\nThis is a brand-new project — created in July 2026 and reaching ~11k stars within four months, with +6,225 in the last tracked month (#18 on the monthly board). That is a spike driven by Tencent's name and a wide feature table, not by a long maintenance record; there is essentially no Hacker News discussion (one thread, 1 point). Treat it as promising but unproven: worth a weekend trial on a spare machine if the multi-user IM story is exactly your gap, not something to put in front of a team yet.",{"slug":25,"name":26},"ai-engineering","AI Engineering",[28,29,30,31,32,33,34],"agent","agentic-ai","ai","ai-agent","ai-agents","local-first","long-term-memory","MIT","https:\u002F\u002Foctop.cloud","2026-10-24T15:37:08Z",[],1792868451]