{"full_name":"vstorm-co/agenticos","owner":"vstorm-co","name":"agenticos","shard":"1","url":"https://github.com/vstorm-co/agenticos","description":"Open-source, self-hosted AI agent platform for your company's AI teammates. Build, run and govern agents on your own servers and models: shared chat, RAG, skills, a real browser, 4,700+ MCP servers, budgets, approvals and an audit trail. An alternative to ChatGPT Enterprise and Grok Bot. Apache-2.0, built on Pydantic AI.","homepage":"https://vstorm-co.github.io/vstormos-landing/","topics":["agentic-ai","agents","pydantic-ai","python","agentic-os","fastapi","local-llm","mcp","on-premise","rag","self-hosted","ai-agents","ai-governance","llm","open-source"],"tag_pages":["agentic-ai","agents","ai-agents","ai-governance","fastapi","llm","local-llm","mcp","on-premise","open-source","python","rag","self-hosted"],"language":"Python","license":"Apache-2.0","avatar":"https://avatars.githubusercontent.com/u/80465946?v=4","created_at":"2026-07-31T09:53:01Z","pushed_at":"2026-11-17T10:20:58Z","metrics":{"stars":385,"forks":67,"watchers":5,"issues":1173,"prs":1868,"releases":701,"issues_open":35,"issues_closed":1138,"prs_open":4,"prs_closed":201,"prs_merged":1663,"discussions":10},"growth":{},"history":{"dates":["2026-11-17"],"stars":[385],"forks":[67],"watchers":[5],"issues_open":[35],"prs_open":[4]},"boards":{},"same_owner":[],"issues":[{"number":168,"title":"Issue map: how the open issues relate, and what blocks what","comments":77,"created_at":"2026-08-04T22:10:25Z"},{"number":2797,"title":"Integrations: built-in Google, Microsoft and Slack on every surface (one PR)","comments":7,"created_at":"2026-11-06T23:16:50Z"},{"number":2927,"title":"Deferred tools must be rediscovered after each chat turn despite appearing in conversation history","comments":3,"created_at":"2026-11-11T21:29:31Z"},{"number":2529,"title":"Check a skill change on every agent of its line, and offer pin or fork when one gets worse","comments":2,"created_at":"2026-10-27T12:58:23Z"},{"number":2526,"title":"Self-improving skills: improve a shared skill across every agent that uses it","comments":2,"created_at":"2026-10-27T12:58:05Z"},{"number":54,"title":"Voice conversations: realtime speech to speech in chat, widget and hosted page","comments":2,"created_at":"2026-08-01T12:49:34Z"},{"number":2555,"title":"Move the evals and Improve screens onto the remaining 1.x components","comments":1,"created_at":"2026-10-28T14:40:16Z"},{"number":2093,"title":"Meetings: web/desktop notes, self-hosted bots and meeting assistant","comments":1,"created_at":"2026-10-10T09:41:02Z"}],"discussions":[{"number":2853,"title":"Kubernetes / Openshift support?","comments":4,"category":"General","created_at":"2026-11-09T21:34:08Z"},{"number":2646,"title":"UI Improvements coming?","comments":1,"category":"General","created_at":"2026-11-02T20:02:45Z"},{"number":2607,"title":"Is podman & podman compose supported?","comments":2,"category":"Q&A","created_at":"2026-11-01T15:03:35Z"},{"number":2594,"title":"Please enroll in anthropics/oss-scanner","comments":1,"category":"General","created_at":"2026-10-31T14:19:02Z"},{"number":2221,"title":"One agent in Slack, Mattermost, Telegram, a widget and the API","comments":0,"category":"General","created_at":"2026-10-18T22:41:16Z"},{"number":2123,"title":"Routines: agents that run on a schedule or an event","comments":0,"category":"General","created_at":"2026-10-12T13:06:04Z"},{"number":2015,"title":"MCP per person: each colleague's own Notion account","comments":0,"category":"General","created_at":"2026-10-05T23:26:29Z"},{"number":2014,"title":"Approvals: sensitive tools wait for a person","comments":1,"category":"General","created_at":"2026-10-05T23:26:28Z"}],"latest_release":{"name":"1.80.0","tag":"v1.80.0","published_at":"2026-11-16T23:01:31Z","body":"### Fixed\n\n- Keep a run alive when an MCP server fails during initialization, discovery or a tool request. Retry initialization once, report unavailable services and uncertain tool outcomes explicitly, and never replay tool writes. Give the built-in platform MCP its own bounded initialization deadline so cold worker startup does not abort Mattermost turns before the model runs (#3077).\n\n- Pin each sync source to its knowledge base instead of resolving its collection name again. Sources, logs and ingested documents stay with the selected base, including same-named personal, organization and deployment bases. Deleting or clearing a base detaches and deactivates exactly its sources; another base does not inherit them. Deleting a base archives its historical source links so its logs cannot fall back to a reassigned source (#2551).\n\n- **Deleting an account.** Deleting a person whose runs belong to their own conversations no longer fails on `agent_runs_conversation_id_fkey`, whichever order the database runs the delete cascades in.\n- **Memory writes after \"forget me\".** A run that began before a person's memory or an agent's memory was erased, one waiting for an approval included, saves nothing more to that memory and tells the model why. Memory extraction holds off clearing an agent's memory while it writes.\n- **Memory-maintenance and daily-briefing manifests.** Erasing a person's memory, or deleting their account, trims the conversation text, facts, notes and requests these runs kept to what the spec decided, including a run that finishes after the erasure.\n- **Task board receipts.** Removing a group, a project or an account purges its task board's table together with every idempotency receipt that holds its records, expired receipts included.\n- **Decision episodes sent to Improve.** **Add as an Improve case** adds the reviewed episode to a dataset you choose, as a case that records who asked it and links its run and conversation, so erasing that person anonymises it like any other case.\n\n- **Workflow approvals reach the people who decide them.** A workflow step asking for approval now notifies its named approvers and the members of its approver groups who hold `approvals:decide`, or, when it names neither, every member of the run's organization holding `approvals:decide`. The notification opens the approvals queue. Pending workflow steps and tool calls the reader may decide lead the Inbox among urgent items, counted with the rest. Approval previews show the action context and requester in compact rows, and a populated Inbox fits its contents instead of stretching an empty panel across the page.\n- **Onboarding offers an owner a model to connect.** An owner who may prepare an agent sees **Connect a model** on the first-task step instead of being told an administrator must finish setting up AI. Onboarding calls the tenant an organization throughout, and a step asked as questions shows one Back, also while it loads or when it fails to.\n- **Dates in the reader's time zone.** Apps, their public pages and every other date the console formats through next-intl follow the browser's time zone, as Chat and Tables do, instead of the server's UTC.\n- **Public link settings appear with the link.** An app's Share dialog shows the link's expiry, version, embedding and password settings only once the link exists, and its address sits in a labelled read-only field.\n- **Forks and the conversation menu.** A forked conversation carries a mark naming its source in the chat list, and the conversation actions are a menu with roles, arrow keys and Escape; Rename keeps the focus in the title field.\n- **Dictation says why it failed.** With the browser's own speech recognition, a blocked microphone, a page that may not ask for one, a missing microphone, a refusing or unreachable speech service and silence each get the step that fixes them, in the console and on hosted chat pages.\n- **One Polish name per concept, and actions in empty lists.** Polish calls a r","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-17T10:29:19Z","history_start":"2026-09-06"}