{"full_name":"agent0ai/agent-zero","owner":"agent0ai","name":"agent-zero","shard":"1","url":"https://github.com/agent0ai/agent-zero","description":"Agent Zero AI framework","homepage":"https://agent-zero.ai","topics":["agent","ai","assistant","autonomous","linux","zero"],"tag_pages":["agent","ai","assistant","linux","zero"],"language":"Python","license":"NOASSERTION","avatar":"https://avatars.githubusercontent.com/u/216033749?v=4","created_at":"2024-06-10T09:10:45Z","pushed_at":"2026-11-14T23:57:09Z","metrics":{"stars":19867,"forks":3878,"watchers":287,"issues":896,"prs":1031,"releases":77,"issues_open":61,"issues_closed":835,"prs_open":111,"prs_closed":601,"prs_merged":319,"discussions":99},"growth":{"day":{"stars":11,"forks":2,"base_day":"2026-11-15"}},"history":{"dates":["2026-11-10","2026-11-11","2026-11-12","2026-11-13","2026-11-14","2026-11-15","2026-11-16"],"stars":[19817,19828,19840,19844,19849,19856,19867],"forks":[3868,3866,3871,3871,3873,3876,3878],"watchers":[null,287,287,287,287,287,287],"issues_open":[null,53,54,54,56,57,61],"prs_open":[null,84,84,84,84,104,111]},"boards":{},"same_owner":[],"issues":[{"number":1847,"title":"Get installation, all set up for agent zero","comments":6,"created_at":"2026-08-22T11:03:12Z"},{"number":1967,"title":"v2.14 + Ollama/Qwen2.5-Coder: repeated or skipped memory_save calls","comments":4,"created_at":"2026-10-22T08:08:53Z"},{"number":1857,"title":"Vision images re-attach as base64 into every LLM request - image-heavy chats hit provider 413 and die silently","comments":4,"created_at":"2026-08-28T13:26:07Z"},{"number":1855,"title":"Chat Completions mode: plain-text final answer is never logged — model answers, turn completes, UI silently shows nothing","comments":4,"created_at":"2026-08-27T11:48:02Z"},{"number":2010,"title":"Self-update to v2.15 rolls back: dependency install times out after 60s","comments":2,"created_at":"2026-11-04T21:00:25Z"},{"number":2003,"title":"code_execution_tool crashes on invalid argument type with local Ollama models (Agent Zero v2.14)","comments":2,"created_at":"2026-11-02T11:45:18Z"},{"number":1889,"title":"Whole-backend wedge: event loop freezes, all chats stall simultaneously, user messages silently swallowed, no trace left","comments":2,"created_at":"2026-09-08T15:42:05Z"},{"number":1840,"title":"Chat click sometimes shows loading splash forever: state_push silently dropped in DEGRADED mode + phantom polling fallback","comments":2,"created_at":"2026-08-19T19:47:34Z"}],"discussions":[{"number":2021,"title":"Task discovery before signup: an inspection-only skill workflow","comments":0,"category":"Show and tell","created_at":"2026-11-07T15:24:02Z"},{"number":1963,"title":"Pinning/Unpinning skills in v2.14","comments":1,"category":"Q&A","created_at":"2026-10-20T11:14:15Z"},{"number":1953,"title":"How do I set the URL for using Ollama in Agent Zero?","comments":1,"category":"Q&A","created_at":"2026-10-14T22:24:01Z"},{"number":1927,"title":"Jev","comments":0,"category":"Q&A","created_at":"2026-09-28T18:51:54Z"},{"number":1885,"title":"JIT Context OS: Running autonomous coding loops on local 9B models with sub-3ms RYOW & zero context bloat (EXP-009)","comments":0,"category":"Show and tell","created_at":"2026-09-08T06:32:07Z"},{"number":1870,"title":"A0 CLI / WebSocket Remote Execution Disconnects After Update — Reproducible Build Blocker","comments":1,"category":"Q&A","created_at":"2026-09-03T21:37:35Z"},{"number":1808,"title":"Agent Only MCP for Agent Zero: shared verified troubleshooting memory","comments":0,"category":"Show and tell","created_at":"2026-08-03T10:16:32Z"},{"number":1795,"title":"Could agent-zero / OpenClaw runs become playable Story artifacts?","comments":0,"category":"Show and tell","created_at":"2026-07-31T02:41:50Z"}],"latest_release":{"name":"v2.15","tag":"v2.15","published_at":"2026-11-01T23:52:00Z","body":"# Agent Zero — Release Notes\n\n## Highlights\n\nThis is a large release that promotes core tools into standalone plugins, overhauls the welcome screen, hardens channel security, improves streaming and UI responsiveness, and adds new search, migration, and documentation capabilities.\n\n---\n\n## Tool Plugin Architecture\n\n- **Extracted bundled tools into standalone plugins:** `wait`, `notify_user`, `search_engine`, `a2a_chat`, `scheduler`, `behavior_adjustment`, and the scheduler Tasks UI are now standard plugins, each with dedicated badge styling, log types, and prompt files.\n- **Unified memory tools:** The four separate `memory_load`, `memory_save`, `memory_delete`, and `memory_forget` tools are consolidated into a single `memory` tool with an `action` argument, matching the action-dispatch pattern used by other tools.\n- **Per-action tool permissions:** Profiles can now allow or deny individual operations across all 12 bundled multi-action tools (e.g., allow file reads but block writes, or allow listing scheduled tasks but block deletion). The Agent Editor shows grouped action controls with expandable entries.\n- **Centralized boolean argument coercion** across all promoted and core tools, removing duplicated helpers and strict-validation error paths.\n- **Dedicated badge rendering** for goal, document_query, skills, vision, browser, office, and computer_use_remote tool calls, each with distinct accent colors and descriptive headlines.\n- **Dropped agent-name prefix** from all tool log headlines, warning rows, and fallback entries for cleaner chat output.\n- **Added catalog thumbnails** for all 47 bundled plugins.\n\n## Output Handling & Model Recovery\n\n- **Surface output-limit truncation instead of misformat retries:** When a turn stops at the token limit without a complete tool request, the agent retries with a dedicated truncation warning instead of the generic misformat message. Fixes #1836.\n- **Guard truncated native calls:** Reject incomplete native-call argument JSON during output-limit recovery while allowing complete calls to proceed. Stream finish metadata is now preserved through early cancellation.\n- **Recover DSML tool calls:** Recognize standalone DSML `function_calls` envelopes, repeated tool names, and outer headlines, routing them through normal dispatch instead of failed conversion.\n- **Keep rejected tool syntax out of model history:** Store neutral `{}` for rejected or incomplete leaked calls so retries don't reinforce malformed output. Replayable Responses output items are removed from the history copy.\n- **Restore Context Doctor repair before retrying malformed calls:** Missing delimiters and other repairable output can execute in the same turn again, restoring pre-PR #1920 behavior.\n- **Keep model prose as thoughts and retry silently:** Free text and reasoning-only output is preserved in the current generation and retried without intermediate warning rows.\n- **Fix include placeholder expansion in tool arguments:** `§§include(path)` is now expanded recursively at the tool execution boundary for Chat, native Responses, and parallel calls.\n- **Explain missing LiteLLM chat streams** with an actionable error instead of raising TypeError.\n\n## WebUI Improvements\n\n- **Faster response streaming:** Reduced state-push coalescing from 100 ms to 25 ms, roughly halving median visible update intervals.\n- **Paint startup overlay immediately** without waiting for render-blocking stylesheets.\n- **Block duplicate sends** while a send is in flight, preventing multiple `/message_async` calls from rapid clicks or Enter presses.\n- **Serialize Welcome chat creation and first send** so rapid input cannot create multiple contexts.\n- **Make composer pause optimistic** against poll snapshots — clicks in flight now suppress snapshot writes to the pause flag.\n- **Keep action buttons mounted during streamed updates** so hover and clicks aren't killed mid-stream.\n- **Preserve pending setup prompts** and retry them automatically after chat preset ch","is_prerelease":false},"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-16T09:23:14Z","history_start":"2026-09-06"}