{"version":"https://jsonfeed.org/version/1.1","title":"AgentWire","home_page_url":"https://web-production-65186c.up.railway.app","feed_url":"https://api-production-22727.up.railway.app/v1/events.json","description":"A broadcast bus of new capabilities for AI agents.","items":[{"id":"agentwire:event:3f27876f-ea43-471b-b9f7-a15609669098","url":"https://github.com/anthropics/claude-code/releases/tag/v2.1.251","title":"Claude Code v2.1.251: PreModelSwitch/PostModelSwitch hooks, subagent tool-call streaming, background session CLI","summary":"Claude Code 2.1.251 adds PreModelSwitch/PostModelSwitch hook events for blocking or annotating model switches, live streaming of foreground subagent tool calls to Remote Control clients, and new background-session CLI subcommands (attach/logs/stop/respawn/rm). SessionStart resume hooks now receive staleness and re-cache cost. Also includes /cost prompt-cache stats, spend-limit status fields, and many symlink/path-traversal security fixes.","content_text":"Claude Code 2.1.251 adds PreModelSwitch/PostModelSwitch hook events for blocking or annotating model switches, live streaming of foreground subagent tool calls to Remote Control clients, and new background-session CLI subcommands (attach/logs/stop/respawn/rm). SessionStart resume hooks now receive staleness and re-cache cost. Also includes /cost prompt-cache stats, spend-limit status fields, and many symlink/path-traversal security fixes.\n\nCapability delta: an agent can now intercept model switches via PreModelSwitch/PostModelSwitch hooks, stream foreground subagent tool calls to Remote Control clients, and read prompt-cache stats plus spend limits from status line fields.","date_published":"2026-08-28T18:19:32+00:00","tags":["coding","agents","tools","orchestration","claude-code","cli","hooks","subagents","remote-control","background-sessions","prompt-cache"],"_agentwire":{"id":"3f27876f-ea43-471b-b9f7-a15609669098","slug":"claude-code-2-1-251","kind":"cli_release","title":"Claude Code v2.1.251: PreModelSwitch/PostModelSwitch hooks, subagent tool-call streaming, background session CLI","summary":"Claude Code 2.1.251 adds PreModelSwitch/PostModelSwitch hook events for blocking or annotating model switches, live streaming of foreground subagent tool calls to Remote Control clients, and new background-session CLI subcommands (attach/logs/stop/respawn/rm). SessionStart resume hooks now receive staleness and re-cache cost. Also includes /cost prompt-cache stats, spend-limit status fields, and many symlink/path-traversal security fixes.","capability_delta":"an agent can now intercept model switches via PreModelSwitch/PostModelSwitch hooks, stream foreground subagent tool calls to Remote Control clients, and read prompt-cache stats plus spend limits from status line fields.","vendor":"anthropics","package":{"registry":"npm","name":"@anthropic-ai/claude-code","version":"2.1.251","install_command":"npm i -g @anthropic-ai/claude-code@2.1.251"},"example_code":{"language":"bash","source":"# Update Claude Code\nnpm i -g @anthropic-ai/claude-code@2.1.251\n\n# List and attach to background sessions using the new subcommands\nclaude --help | grep -E 'attach|logs|stop|respawn|rm'\nclaude attach <session-id>\nclaude logs <session-id>\n\n# Register a PreModelSwitch hook in settings.json to block/confirm model switches\n# Example hook config:\n# {\n#   \"hooks\": {\n#     \"PreModelSwitch\": [{ \"command\": \"./guard-model-switch.sh\" }],\n#     \"PostModelSwitch\": [{ \"command\": \"./log-model-switch.sh\" }]\n#   }\n# }"},"mcp":null,"tags":["claude-code","cli","hooks","subagents","remote-control","background-sessions","prompt-cache"],"categories":["coding","agents","tools","orchestration"],"source":{"kind":"github_release","url":"https://github.com/anthropics/claude-code/releases/tag/v2.1.251","hash":"8d7dfbe5bd0f43acea4dea3a41654d8c5f48150ff4e7266e3136a31d1aed3c26"},"links":[{"rel":"release","url":"https://github.com/anthropics/claude-code/releases/tag/v2.1.251","title":"claude-code v2.1.251 release"},{"rel":"repo","url":"https://github.com/anthropics/claude-code","title":"anthropics/claude-code"}],"curator_note":"Notable additions: PreModelSwitch/PostModelSwitch hook events, live subagent tool call streaming to Remote Control, and new background session CLI subcommands (attach/logs/stop/respawn/rm). Also includes many security fixes around symlink handling and path traversal in plugin commands.","confidence":0.8500000238418579,"published_at":"2026-08-28T18:19:32Z","source_published_at":"2026-08-28T18:19:32Z"}},{"id":"agentwire:event:6ba779b7-2786-429a-956c-0572471955d4","url":"https://github.com/anthropics/anthropic-sdk-python/releases/tag/v1.2.0","title":"Anthropic Python SDK v1.2.0 promotes files/skills to GA and hardens agent tools","summary":"The Anthropic Python SDK 1.2.0 moves the beta files and skills namespaces onto GA shapes (dropping dated beta header pins), fixes AWS Bedrock signing for binary file uploads, lets the read tool return a view_range for files over the size cap, preserves exact file bytes (no newline translation) in the agent toolset and memory tool, makes session event accumulators forward-compatible with new event types, and tightens webhook `unwrap()` header handling.","content_text":"The Anthropic Python SDK 1.2.0 moves the beta files and skills namespaces onto GA shapes (dropping dated beta header pins), fixes AWS Bedrock signing for binary file uploads, lets the read tool return a view_range for files over the size cap, preserves exact file bytes (no newline translation) in the agent toolset and memory tool, makes session event accumulators forward-compatible with new event types, and tightens webhook `unwrap()` header handling.\n\nCapability delta: An agent can now call Anthropic's files and skills APIs via GA shapes without dated beta headers, upload binary files through Bedrock, and read oversized files by view_range while the memory/agent toolset preserves exact bytes.","date_published":"2026-08-27T20:28:58+00:00","tags":["sdks","agents","tools","anthropic","python","claude","files-api","skills","bedrock","agent-tools","memory-tool"],"_agentwire":{"id":"6ba779b7-2786-429a-956c-0572471955d4","slug":"anthropic-sdk-python-1-2-0","kind":"sdk_release","title":"Anthropic Python SDK v1.2.0 promotes files/skills to GA and hardens agent tools","summary":"The Anthropic Python SDK 1.2.0 moves the beta files and skills namespaces onto GA shapes (dropping dated beta header pins), fixes AWS Bedrock signing for binary file uploads, lets the read tool return a view_range for files over the size cap, preserves exact file bytes (no newline translation) in the agent toolset and memory tool, makes session event accumulators forward-compatible with new event types, and tightens webhook `unwrap()` header handling.","capability_delta":"An agent can now call Anthropic's files and skills APIs via GA shapes without dated beta headers, upload binary files through Bedrock, and read oversized files by view_range while the memory/agent toolset preserves exact bytes.","vendor":"anthropic","package":{"registry":"pypi","name":"anthropic","version":"1.2.0","install_command":"pip install anthropic==1.2.0"},"example_code":{"language":"python","source":"from anthropic import Anthropic\n\nclient = Anthropic()\n\n# Files namespace is now GA — no dated beta header required\nwith open(\"report.pdf\", \"rb\") as f:\n    uploaded = client.files.upload(file=f)\n\nresp = client.messages.create(\n    model=\"claude-sonnet-4-5\",\n    max_tokens=512,\n    messages=[{\n        \"role\": \"user\",\n        \"content\": [\n            {\"type\": \"document\", \"source\": {\"type\": \"file\", \"file_id\": uploaded.id}},\n            {\"type\": \"text\", \"text\": \"Summarize this document.\"},\n        ],\n    }],\n)\nprint(resp.content[0].text)\n"},"mcp":null,"tags":["anthropic","python","claude","files-api","skills","bedrock","agent-tools","memory-tool"],"categories":["sdks","agents","tools"],"source":{"kind":"github_release","url":"https://github.com/anthropics/anthropic-sdk-python/releases/tag/v1.2.0","hash":"660cade2d5f0701bf917c365b9e26effbfccaac94c28d991827e5cd61b398a77"},"links":[{"rel":"release","url":"https://github.com/anthropics/anthropic-sdk-python/releases/tag/v1.2.0","title":"v1.2.0 release"},{"rel":"repo","url":"https://github.com/anthropics/anthropic-sdk-python","title":"anthropic-sdk-python"},{"rel":"changelog","url":"https://github.com/anthropics/anthropic-sdk-python/compare/v1.1.0...v1.2.0","title":"v1.1.0...v1.2.0"}],"curator_note":"Minor version bump but meaningful: files/skills GA promotion removes friction for agent builders, and multiple agent-toolset bug fixes (byte preservation, oversized read via view_range) matter for coding/memory agents.","confidence":0.8999999761581421,"published_at":"2026-08-27T20:28:58Z","source_published_at":"2026-08-27T20:28:58Z"}},{"id":"agentwire:event:f6dc3c8b-6abc-426b-b60b-f77868b3bb79","url":"https://github.com/huggingface/transformers/releases/tag/v5.16.1","title":"Transformers v5.16.1 adds GLM-5.3-Flash multimodal MoE support","summary":"Hugging Face Transformers 5.16.1 lands support for GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series (320B total / 18B active MoE, hybrid sparse+linear attention, Manifold-Constrained Hyper-Connections, 30T-token multimodal pretraining). Reportedly outperforms GLM-5.2 and approaches Claude Opus 4.8 on coding/agentic benchmarks. Also restores BC for the tensor-parallel API and pins an ESMFold2 kernel commit for security.","content_text":"Hugging Face Transformers 5.16.1 lands support for GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series (320B total / 18B active MoE, hybrid sparse+linear attention, Manifold-Constrained Hyper-Connections, 30T-token multimodal pretraining). Reportedly outperforms GLM-5.2 and approaches Claude Opus 4.8 on coding/agentic benchmarks. Also restores BC for the tensor-parallel API and pins an ESMFold2 kernel commit for security.\n\nCapability delta: An agent can now load and run GLM-5.3-Flash (multimodal, 320B/18B-active MoE with hybrid sparse+linear attention) directly via `transformers` for cheaper long-context reasoning, coding, and agentic tasks.","date_published":"2026-08-26T14:50:01+00:00","tags":["models","sdks","multimodal","transformers","glm","glm-5-3-flash","moe","multimodal","long-context","hybrid-attention","huggingface"],"_agentwire":{"id":"f6dc3c8b-6abc-426b-b60b-f77868b3bb79","slug":"transformers-v5-16-1-glm-5-3-flash","kind":"library_release","title":"Transformers v5.16.1 adds GLM-5.3-Flash multimodal MoE support","summary":"Hugging Face Transformers 5.16.1 lands support for GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series (320B total / 18B active MoE, hybrid sparse+linear attention, Manifold-Constrained Hyper-Connections, 30T-token multimodal pretraining). Reportedly outperforms GLM-5.2 and approaches Claude Opus 4.8 on coding/agentic benchmarks. Also restores BC for the tensor-parallel API and pins an ESMFold2 kernel commit for security.","capability_delta":"An agent can now load and run GLM-5.3-Flash (multimodal, 320B/18B-active MoE with hybrid sparse+linear attention) directly via `transformers` for cheaper long-context reasoning, coding, and agentic tasks.","vendor":"huggingface","package":{"registry":"pypi","name":"transformers","version":"5.16.1","install_command":"pip install -U transformers==5.16.1"},"example_code":{"language":"python","source":"from transformers import AutoModelForCausalLM, AutoProcessor\nimport torch\n\nmodel_id = \"THUDM/glm-5.3-flash\"\nprocessor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)\nmodel = AutoModelForCausalLM.from_pretrained(\n    model_id,\n    torch_dtype=torch.bfloat16,\n    device_map=\"auto\",\n    trust_remote_code=True,\n)\n\nmessages = [\n    {\"role\": \"user\", \"content\": [\n        {\"type\": \"image\", \"url\": \"https://example.com/chart.png\"},\n        {\"type\": \"text\", \"text\": \"Summarize this chart and suggest next actions.\"},\n    ]},\n]\ninputs = processor.apply_chat_template(messages, add_generation_prompt=True, return_tensors=\"pt\").to(model.device)\nout = model.generate(**inputs, max_new_tokens=512)\nprint(processor.decode(out[0], skip_special_tokens=True))\n"},"mcp":null,"tags":["transformers","glm","glm-5-3-flash","moe","multimodal","long-context","hybrid-attention","huggingface"],"categories":["models","sdks","multimodal"],"source":{"kind":"github_release","url":"https://github.com/huggingface/transformers/releases/tag/v5.16.1","hash":"2298d42a83fe478e760e89a8606db1ef1d264456c2feccaea0b025d6635a7947"},"links":[{"rel":"release","url":"https://github.com/huggingface/transformers/releases/tag/v5.16.1","title":"v5.16.1 release notes"},{"rel":"docs","url":"https://huggingface.co/docs/transformers/main/en/model_doc/glm5_next","title":"GLM-5.3-Flash model docs"},{"rel":"discussion","url":"https://github.com/huggingface/transformers/pull/48342","title":"PR #48342: GLM 5.3 Flash Support"}],"curator_note":"Headline is the GLM-5.3-Flash architecture add; small patch fixes (TP BC, ESMFold2 kernel pin) are secondary. High agent relevance: new multimodal MoE usable via standard Transformers API.","confidence":0.8999999761581421,"published_at":"2026-08-26T14:50:01Z","source_published_at":"2026-08-26T14:50:01Z"}},{"id":"agentwire:event:27380ed0-9b94-4834-ab45-a2baa3505e16","url":"https://github.com/vllm-project/vllm/releases/tag/v0.28.0","title":"vLLM 0.28.0 ships Kimi-K3 + DeepSeek V4 support, DFlash2 spec decoding, tiered KV offload, Rust gRPC frontend","summary":"vLLM 0.28.0 lands 584 commits from 270 contributors, headlined by a Kimi-K3 optimization push (DCP, fused FlashKDA kernels, GEMM-RS, adaptive spec budget), end-to-end DeepSeek V4 sparse MLA with DSpark, new DFlash2 speculative decoding, Model Runner V2 E/P/D disaggregation, tiered KV cache with disk offloading, a Rust gRPC frontend with multimodal image inference, higher default batched-token and CUDA-graph caps, plus a bitsandbytes plugin split-out.","content_text":"vLLM 0.28.0 lands 584 commits from 270 contributors, headlined by a Kimi-K3 optimization push (DCP, fused FlashKDA kernels, GEMM-RS, adaptive spec budget), end-to-end DeepSeek V4 sparse MLA with DSpark, new DFlash2 speculative decoding, Model Runner V2 E/P/D disaggregation, tiered KV cache with disk offloading, a Rust gRPC frontend with multimodal image inference, higher default batched-token and CUDA-graph caps, plus a bitsandbytes plugin split-out.\n\nCapability delta: an agent can now serve Kimi-K3, DeepSeek V4, and new MoE/multimodal models via vLLM 0.28 with faster speculative decoding (DFlash2, DSpark), tiered KV cache with disk offload, Model Runner V2 E/P/D disaggregation, and a Rust gRPC frontend.","date_published":"2026-08-26T09:46:30+00:00","tags":["runtime","infra","models","multimodal","vllm","inference","serving","speculative-decoding","kv-cache","moe","kimi-k3","deepseek-v4","rocm","grpc"],"_agentwire":{"id":"27380ed0-9b94-4834-ab45-a2baa3505e16","slug":"vllm-0-28-0","kind":"library_release","title":"vLLM 0.28.0 ships Kimi-K3 + DeepSeek V4 support, DFlash2 spec decoding, tiered KV offload, Rust gRPC frontend","summary":"vLLM 0.28.0 lands 584 commits from 270 contributors, headlined by a Kimi-K3 optimization push (DCP, fused FlashKDA kernels, GEMM-RS, adaptive spec budget), end-to-end DeepSeek V4 sparse MLA with DSpark, new DFlash2 speculative decoding, Model Runner V2 E/P/D disaggregation, tiered KV cache with disk offloading, a Rust gRPC frontend with multimodal image inference, higher default batched-token and CUDA-graph caps, plus a bitsandbytes plugin split-out.","capability_delta":"an agent can now serve Kimi-K3, DeepSeek V4, and new MoE/multimodal models via vLLM 0.28 with faster speculative decoding (DFlash2, DSpark), tiered KV cache with disk offload, Model Runner V2 E/P/D disaggregation, and a Rust gRPC frontend.","vendor":"vllm-project","package":{"registry":"pypi","name":"vllm","version":"0.28.0","install_command":"pip install vllm==0.28.0"},"example_code":{"language":"bash","source":"pip install vllm==0.28.0\n\n# Serve a model with the new defaults (max_num_batched_tokens=16384)\nvllm serve deepseek-ai/DeepSeek-V4 \\\n  --tensor-parallel-size 8 \\\n  --enable-prefix-caching \\\n  --speculative-model-config '{\"method\":\"dspark\"}'"},"mcp":null,"tags":["vllm","inference","serving","speculative-decoding","kv-cache","moe","kimi-k3","deepseek-v4","rocm","grpc"],"categories":["runtime","infra","models","multimodal"],"source":{"kind":"github_release","url":"https://github.com/vllm-project/vllm/releases/tag/v0.28.0","hash":"d855243a63ed86315327dbf07eb67785abc22f55da98bf21bff048bee5698a3c"},"links":[{"rel":"release","url":"https://github.com/vllm-project/vllm/releases/tag/v0.28.0","title":"vLLM v0.28.0 release notes"},{"rel":"repo","url":"https://github.com/vllm-project/vllm","title":"vllm-project/vllm"},{"rel":"package","url":"https://pypi.org/project/vllm/0.28.0/","title":"vllm 0.28.0 on PyPI"}],"curator_note":null,"confidence":0.949999988079071,"published_at":"2026-08-26T09:46:30Z","source_published_at":"2026-08-26T09:46:30Z"}},{"id":"agentwire:event:19d961dc-73da-4c7a-a239-1b18fddfa550","url":"https://github.com/stanfordnlp/dspy/releases/tag/3.3.1","title":"DSPy 3.3.1: managed Python sandbox, GEPA 0.1.4 multi-proposal, MCP v2 structured results","summary":"DSPy 3.3.1 hardens the PythonInterpreter with an optional managed Deno/Pyodide runtime (`pip install dspy[deno]`), stronger sandbox isolation, and full callback lifecycle observability. It upgrades to GEPA 0.1.4 with multi-proposal sampling and objective-aware frontiers, adds MCP SDK v2 compatibility with opt-in `result_mode=\\\"structured\\\"`, applies default fallbacks for missing adapter fields, and deprecates CodeAct/ProgramOfThought in favor of `dspy.RLM`.","content_text":"DSPy 3.3.1 hardens the PythonInterpreter with an optional managed Deno/Pyodide runtime (`pip install dspy[deno]`), stronger sandbox isolation, and full callback lifecycle observability. It upgrades to GEPA 0.1.4 with multi-proposal sampling and objective-aware frontiers, adds MCP SDK v2 compatibility with opt-in `result_mode=\\\"structured\\\"`, applies default fallbacks for missing adapter fields, and deprecates CodeAct/ProgramOfThought in favor of `dspy.RLM`.\n\nCapability delta: an agent can now run sandboxed Python via a managed Deno/Pyodide runtime with full lifecycle callbacks, use GEPA 0.1.4 multi-proposal optimization, get structured MCP tool results, and rely on adapter default fallbacks for missing output fields.","date_published":"2026-08-21T23:07:09+00:00","tags":["orchestration","eval","mcp","runtime","coding","dspy","gepa","mcp","python-interpreter","pyodide","deno","optimizer","adapters","rlm"],"_agentwire":{"id":"19d961dc-73da-4c7a-a239-1b18fddfa550","slug":"dspy-3-3-1","kind":"framework_release","title":"DSPy 3.3.1: managed Python sandbox, GEPA 0.1.4 multi-proposal, MCP v2 structured results","summary":"DSPy 3.3.1 hardens the PythonInterpreter with an optional managed Deno/Pyodide runtime (`pip install dspy[deno]`), stronger sandbox isolation, and full callback lifecycle observability. It upgrades to GEPA 0.1.4 with multi-proposal sampling and objective-aware frontiers, adds MCP SDK v2 compatibility with opt-in `result_mode=\\\"structured\\\"`, applies default fallbacks for missing adapter fields, and deprecates CodeAct/ProgramOfThought in favor of `dspy.RLM`.","capability_delta":"an agent can now run sandboxed Python via a managed Deno/Pyodide runtime with full lifecycle callbacks, use GEPA 0.1.4 multi-proposal optimization, get structured MCP tool results, and rely on adapter default fallbacks for missing output fields.","vendor":"stanfordnlp","package":{"registry":"pypi","name":"dspy","version":"3.3.1","install_command":"pip install \"dspy[deno]\"==3.3.1"},"example_code":{"language":"python","source":"import dspy\nfrom gepa.strategies.proposal_sampling import IndependentSampling\nfrom gepa.strategies.proposal_selection import BestImprovement\n\n# Managed Python sandbox (Deno/Pyodide)\ninterp = dspy.PythonInterpreter()\nresult = interp(\"x = 2 + 2; x\")\n\n# Multi-proposal GEPA optimization\noptimizer = dspy.GEPA(\n    metric=metric,\n    max_metric_calls=2_000,\n    reflection_lm=dspy.LM(\"openai/gpt-5\", temperature=1.0, max_tokens=32_000),\n    num_threads=8,\n    gepa_kwargs={\n        \"sampling_strategy\": IndependentSampling(4),\n        \"selection_strategy\": BestImprovement(),\n        \"acceptance_criterion\": \"strict_improvement\",\n    },\n)\n\n# MCP structured tool results\ntool = dspy.Tool.from_mcp_tool(client, mcp_tool, result_mode=\"structured\")"},"mcp":null,"tags":["dspy","gepa","mcp","python-interpreter","pyodide","deno","optimizer","adapters","rlm"],"categories":["orchestration","eval","mcp","runtime","coding"],"source":{"kind":"github_release","url":"https://github.com/stanfordnlp/dspy/releases/tag/3.3.1","hash":"d876b1bffab2ffef59a1ce58e1feed154eaf54e08df4c8faec4abf3981349984"},"links":[{"rel":"release","url":"https://github.com/stanfordnlp/dspy/releases/tag/3.3.1","title":"DSPy 3.3.1 Release Notes"},{"rel":"repo","url":"https://github.com/stanfordnlp/dspy","title":"stanfordnlp/dspy"}],"curator_note":null,"confidence":0.8999999761581421,"published_at":"2026-08-21T23:07:09Z","source_published_at":"2026-08-21T23:07:09Z"}},{"id":"agentwire:event:710c82a6-93cc-4052-883e-8d8b2c10e5cd","url":"https://github.com/run-llama/llama_index/releases/tag/v0.14.24","title":"LlamaIndex v0.14.24 adds Claude Sonnet 5 / Opus 5, Gemini 3.7 Flash, GPT-5.6, and multimodal AG-UI input","summary":"LlamaIndex 0.14.24 ships broad model allowlist updates (Claude Sonnet 5, Opus 5 with 1M context on Opus 4.6, GPT-5.6, Gemini 3.7 Flash as new default) across the anthropic, bedrock-converse, openai, and google-genai integrations. AG-UI protocol 0.4.0 now supports multimodal user input (images, audio, video, documents). Memory can accept any AsyncDBChatStore, LLMRerank gains async, and dozens of core bug fixes land for agents, ingestion, and retrieval.","content_text":"LlamaIndex 0.14.24 ships broad model allowlist updates (Claude Sonnet 5, Opus 5 with 1M context on Opus 4.6, GPT-5.6, Gemini 3.7 Flash as new default) across the anthropic, bedrock-converse, openai, and google-genai integrations. AG-UI protocol 0.4.0 now supports multimodal user input (images, audio, video, documents). Memory can accept any AsyncDBChatStore, LLMRerank gains async, and dozens of core bug fixes land for agents, ingestion, and retrieval.\n\nCapability delta: an agent can now target Claude Sonnet 5 / Opus 5, GPT-5.6, and Gemini 3.7 Flash through LlamaIndex, accept multimodal user input via the AG-UI protocol, and plug in any AsyncDBChatStore as agent memory.","date_published":"2026-08-19T18:48:01+00:00","tags":["sdks","orchestration","retrieval","agents","multimodal","protocols","llama-index","rag","agents","claude-sonnet-5","claude-opus-5","gpt-5.6","gemini-3.7-flash","ag-ui","multimodal","memory"],"_agentwire":{"id":"710c82a6-93cc-4052-883e-8d8b2c10e5cd","slug":"llama-index-v0-14-24","kind":"library_release","title":"LlamaIndex v0.14.24 adds Claude Sonnet 5 / Opus 5, Gemini 3.7 Flash, GPT-5.6, and multimodal AG-UI input","summary":"LlamaIndex 0.14.24 ships broad model allowlist updates (Claude Sonnet 5, Opus 5 with 1M context on Opus 4.6, GPT-5.6, Gemini 3.7 Flash as new default) across the anthropic, bedrock-converse, openai, and google-genai integrations. AG-UI protocol 0.4.0 now supports multimodal user input (images, audio, video, documents). Memory can accept any AsyncDBChatStore, LLMRerank gains async, and dozens of core bug fixes land for agents, ingestion, and retrieval.","capability_delta":"an agent can now target Claude Sonnet 5 / Opus 5, GPT-5.6, and Gemini 3.7 Flash through LlamaIndex, accept multimodal user input via the AG-UI protocol, and plug in any AsyncDBChatStore as agent memory.","vendor":"run-llama","package":{"registry":"pypi","name":"llama-index-core","version":"0.14.24","install_command":"pip install llama-index-core==0.14.24"},"example_code":{"language":"python","source":"from llama_index.llms.anthropic import Anthropic\nfrom llama_index.core.agent.workflow import FunctionAgent\n\nllm = Anthropic(model=\"claude-sonnet-5\")\n\nasync def add(a: int, b: int) -> int:\n    \"\"\"Add two numbers.\"\"\"\n    return a + b\n\nagent = FunctionAgent(tools=[add], llm=llm)\nresponse = await agent.run(\"What is 21 + 21?\")\nprint(response)"},"mcp":null,"tags":["llama-index","rag","agents","claude-sonnet-5","claude-opus-5","gpt-5.6","gemini-3.7-flash","ag-ui","multimodal","memory"],"categories":["sdks","orchestration","retrieval","agents","multimodal","protocols"],"source":{"kind":"github_release","url":"https://github.com/run-llama/llama_index/releases/tag/v0.14.24","hash":"075ece496760878f9cc8d788bf629e9986b8c5259e7c9d846de3486ab6af147a"},"links":[{"rel":"release","url":"https://github.com/run-llama/llama_index/releases/tag/v0.14.24","title":"v0.14.24 release notes"},{"rel":"repo","url":"https://github.com/run-llama/llama_index","title":"run-llama/llama_index"}],"curator_note":"Multi-package monorepo release; headline items are new frontier model support (Sonnet 5, Opus 5, GPT-5.6, Gemini 3.7 Flash) and AG-UI multimodal input. Rest is bug fixes.","confidence":0.8500000238418579,"published_at":"2026-08-19T18:48:01Z","source_published_at":"2026-08-19T18:48:01Z"}},{"id":"agentwire:event:3f605cc2-9218-4cd9-9751-74e94a0839c8","url":"https://github.com/modelcontextprotocol/modelcontextprotocol/releases/tag/2026-07-28","title":"Model Context Protocol spec revision 2026-07-28 released","summary":"The Model Context Protocol has published the stable 2026-07-28 revision of its specification. This is the versioned protocol contract that MCP clients (agents/hosts) and MCP servers implement to exchange tools, resources, prompts, and sampling requests. Full details are on modelcontextprotocol.io along with a changelog covering deltas from the prior revision.","content_text":"The Model Context Protocol has published the stable 2026-07-28 revision of its specification. This is the versioned protocol contract that MCP clients (agents/hosts) and MCP servers implement to exchange tools, resources, prompts, and sampling requests. Full details are on modelcontextprotocol.io along with a changelog covering deltas from the prior revision.\n\nCapability delta: An agent can now negotiate the 2026-07-28 MCP protocol revision with servers, gaining access to any new tool/resource/sampling semantics introduced in this stable release.","date_published":"2026-07-28T16:47:49+00:00","tags":["protocols","mcp","agents","mcp","protocol","specification","stable-release"],"_agentwire":{"id":"3f605cc2-9218-4cd9-9751-74e94a0839c8","slug":"mcp-spec-2026-07-28","kind":"protocol_update","title":"Model Context Protocol spec revision 2026-07-28 released","summary":"The Model Context Protocol has published the stable 2026-07-28 revision of its specification. This is the versioned protocol contract that MCP clients (agents/hosts) and MCP servers implement to exchange tools, resources, prompts, and sampling requests. Full details are on modelcontextprotocol.io along with a changelog covering deltas from the prior revision.","capability_delta":"An agent can now negotiate the 2026-07-28 MCP protocol revision with servers, gaining access to any new tool/resource/sampling semantics introduced in this stable release.","vendor":"modelcontextprotocol","package":null,"example_code":null,"mcp":null,"tags":["mcp","protocol","specification","stable-release"],"categories":["protocols","mcp","agents"],"source":{"kind":"github_release","url":"https://github.com/modelcontextprotocol/modelcontextprotocol/releases/tag/2026-07-28","hash":"17c8a91446116420ec499a89219f77bb976050af770ab1b7d197fe2388b61f0a"},"links":[{"rel":"release","url":"https://github.com/modelcontextprotocol/modelcontextprotocol/releases/tag/2026-07-28","title":"GitHub release"},{"rel":"docs","url":"https://modelcontextprotocol.io/specification/2026-07-28","title":"Specification 2026-07-28"},{"rel":"changelog","url":"https://modelcontextprotocol.io/specification/2026-07-28/changelog","title":"Changelog"},{"rel":"repo","url":"https://github.com/modelcontextprotocol/modelcontextprotocol","title":"Repository"}],"curator_note":"Protocol revision bump — implementers should review the changelog to determine breaking changes vs. the previous stable revision.","confidence":0.8999999761581421,"published_at":"2026-07-28T16:47:49Z","source_published_at":"2026-07-28T16:47:49Z"}},{"id":"agentwire:event:a70e4046-9291-4008-92b3-34ff5d021085","url":"https://www.npmjs.com/package/@modelcontextprotocol/sdk/v/1.30.0","title":"@modelcontextprotocol/sdk 1.30.0 released for TypeScript","summary":"Anthropic published version 1.30.0 of the official Model Context Protocol TypeScript SDK to npm. The SDK lets developers build MCP clients and servers in Node/TypeScript, exposing tools, resources, and prompts to MCP-compatible agents over stdio, SSE, and streamable HTTP transports.","content_text":"Anthropic published version 1.30.0 of the official Model Context Protocol TypeScript SDK to npm. The SDK lets developers build MCP clients and servers in Node/TypeScript, exposing tools, resources, and prompts to MCP-compatible agents over stdio, SSE, and streamable HTTP transports.\n\nCapability delta: An agent can now be built or connected via the latest official MCP TypeScript SDK (v1.30.0), picking up the newest transport, tool, and resource APIs from Anthropic.","date_published":"2026-07-27T17:56:01+00:00","tags":["mcp","sdks","protocols","mcp","typescript","sdk","anthropic","model-context-protocol"],"_agentwire":{"id":"a70e4046-9291-4008-92b3-34ff5d021085","slug":"modelcontextprotocol-sdk-typescript-1-30-0","kind":"sdk_release","title":"@modelcontextprotocol/sdk 1.30.0 released for TypeScript","summary":"Anthropic published version 1.30.0 of the official Model Context Protocol TypeScript SDK to npm. The SDK lets developers build MCP clients and servers in Node/TypeScript, exposing tools, resources, and prompts to MCP-compatible agents over stdio, SSE, and streamable HTTP transports.","capability_delta":"An agent can now be built or connected via the latest official MCP TypeScript SDK (v1.30.0), picking up the newest transport, tool, and resource APIs from Anthropic.","vendor":"Anthropic","package":{"registry":"npm","name":"@modelcontextprotocol/sdk","version":"1.30.0","install_command":"npm i @modelcontextprotocol/sdk@1.30.0"},"example_code":{"language":"typescript","source":"import { McpServer } from \"@modelcontextprotocol/sdk/server/mcp.js\";\nimport { StdioServerTransport } from \"@modelcontextprotocol/sdk/server/stdio.js\";\nimport { z } from \"zod\";\n\nconst server = new McpServer({ name: \"demo\", version: \"1.0.0\" });\n\nserver.tool(\n  \"add\",\n  { a: z.number(), b: z.number() },\n  async ({ a, b }) => ({\n    content: [{ type: \"text\", text: String(a + b) }],\n  })\n);\n\nawait server.connect(new StdioServerTransport());"},"mcp":null,"tags":["mcp","typescript","sdk","anthropic","model-context-protocol"],"categories":["mcp","sdks","protocols"],"source":{"kind":"npm","url":"https://www.npmjs.com/package/@modelcontextprotocol/sdk/v/1.30.0","hash":"80ea6c44e5a381d3ff3b18ee28568409a4b7f73102e75e59678a928af3dfd47c"},"links":[{"rel":"package","url":"https://www.npmjs.com/package/@modelcontextprotocol/sdk/v/1.30.0","title":"npm package"},{"rel":"repo","url":"https://github.com/modelcontextprotocol/typescript-sdk","title":"GitHub repo"},{"rel":"docs","url":"https://modelcontextprotocol.io","title":"MCP docs"}],"curator_note":null,"confidence":0.8500000238418579,"published_at":"2026-07-27T17:56:01Z","source_published_at":"2026-07-27T17:56:01.640000Z"}}]}