# FlyMyAI Headless Cloud - platform log

Current version: **1.36.0** (2026-11-10). Newest first; this page shows the latest 10 entries.

- What the platform does today (the current snapshot): https://flymy.ai/skill.md
- Every entry, machine-readable: https://flymy.ai/changelog.json
- Agents: call `whats_new` on the Agents MCP with `since` set to the last version you saw.

## 1.36.0 - 2026-11-10 - Long generations as tasks: run_model answers at once, get_model_result reads the result

- **models** - run_model no longer holds the call for a whole video. A generation that takes longer than wait_ms (25 seconds unless you set it) answers status "running" with a prediction_id while it goes on, and get_model_result reads the result when it is ready. Reading is free and safe to repeat. Use: `run_model`, `get_model_result`. [Docs](https://docs.flymy.ai/agents/guides/generate-media-from-claude)
  Try: "Generate a 5-second video of a paper plane flying through a neon city and show it when it is ready."
- **mcp** - An assistant can start several generations and collect them as they finish. wait_ms sets how long run_model waits before it hands the generation back; timeout_ms without wait_ms keeps the older call that waits for the result. Use: `run_model`, `get_model_result`. [Docs](https://docs.flymy.ai/agents/guides/agent-gateway-reference)
  Try: "Start three image variations of my logo at once and show each one as it finishes."

## 1.35.4 - 2026-11-10 - Read the Hacker News front-page Atom feed

- **tools** - Read the current Hacker News front-page Atom feed as complete XML with your agent, without a provider account. Extract story links and feed metadata from the returned feed_xml. Use: `search_tools`, `get_tool_schemas`, `execute_tool`. [Docs](https://docs.flymy.ai/mcp/hackernews#hackernews-get-front-page-feed)
  Try: "Fetch the Hacker News front-page Atom feed and list its story links."

## 1.35.3 - 2026-11-10 - Read and search Hacker News without a provider account

- **tools** - Connect Hacker News without a provider account to read stories, comments and public profiles, browse feeds, and search discussions with filters. Use: `search_tools`, `get_tool_schemas`, `execute_tool`. [Docs](https://docs.flymy.ai/mcp/hackernews)
  Try: "Show me Hacker News front-page stories with at least 40 points."

## 1.35.2 - 2026-11-09 - ElevenLabs music, multi-voice dialogue and Eleven v4 voices

- **tools** - ElevenLabs composes music: compose_music writes a song or an instrumental from a prompt or from a composition plan you can edit, and video_to_music scores background music to your videos. Eleven Music costs $0.15 per minute of music. Use: `compose_music`, `music_composition_plan`, `video_to_music`. [Docs](https://docs.flymy.ai/mcp/elevenlabs)
  Try: "Compose a 30-second upbeat instrumental for my product video with ElevenLabs."
- **tools** - text_to_dialogue voices a conversation between up to 10 ElevenLabs voices in one audio file and, on request, says when each word is spoken and by whom. Use: `text_to_dialogue`, `get_voices`. [Docs](https://docs.flymy.ai/mcp/elevenlabs)
  Try: "Make a 20-second podcast intro where two hosts greet each other, with ElevenLabs dialogue."
- **tools** - ElevenLabs speech runs on Eleven v4 and v4 Turbo and can return each word's start and end for subtitles, at ElevenLabs' $0.08 per 1,000 characters ($0.04 on v4 Turbo and Flash). Use: `text_to_speech`. [Docs](https://docs.flymy.ai/mcp/elevenlabs)
  Try: "Read my product announcement aloud with ElevenLabs v4 and give me subtitle timings."
- **tools** - forced_alignment times a transcript you already have against its audio, word by word, and remix_voice changes the accent, age or tone of a voice you designed. Use: `forced_alignment`, `remix_voice`, `create_voice_from_preview`. [Docs](https://docs.flymy.ai/mcp/elevenlabs)
  Try: "Time this transcript against my podcast audio so I can make subtitles."

## 1.35.1 - 2026-11-09 - JSON answers under your schema from chat models

- **api** - The OpenAI-compatible API takes response_format: with json_schema a chat model answers JSON under your schema, with json_object any JSON object. It works on Gemini 2.5 Flash, 3.1 Flash Lite and 3.1 Pro, Grok 4.7, DeepSeek V4.1 Flash and Kimi K3.1 Pro. Use: `/api/v1/{owner}/{model}/openai/v1/chat/completions`. [Docs](https://docs.flymy.ai/guides/llm-openai-compat#structured-output)
  Try: "Show me how to get invoice fields as JSON under my schema from Gemini 2.5 Flash with the OpenAI SDK."

## 1.35.0 - 2026-11-08 - Stop a helper early; Google results with their own pages

- **agents** - A lead can act before its whole batch of helpers ends: delegate_agent or delegate with wake_on_end=true wakes it as soon as that helper ends, so it uses the result and stops a helper it no longer needs with cancel_subagents(child_id). Without it the batch joins as before. Use: `run_agent`, `get_run`. [Docs](https://docs.flymy.ai/agents/guides/serverless-agents)
  Try: "Start two research helpers on the same question and stop the slower one as soon as the first answers."
- **tools** - DataForSEO's Standard queue: dataforseo_serp_google_organic_standard_post queues up to 100 Google searches in one call ($0.06 for 100 at 10 results) and standard_get reads them free. Every organic result comes with its own page; live results often come as google.com/goto links. Use: `search_tools`, `execute_tool`. [Docs](https://docs.flymy.ai/mcp/dataforseo)
  Try: "Check where my site ranks on Google for these 50 searches through the DataForSEO queue."
- **agents** - update_agent that only removes a tool, server, resource set or skill saves no new version: the removal already applies to every version, so there is nothing to test or make live. Use: `update_agent`. [Docs](https://docs.flymy.ai/agents/guides/serverless-agents)
  Try: "Take Gmail away from my scheduled agent."

## 1.34.0 - 2026-11-08 - A frozen version keeps its model and tools

- **agents** - A frozen version keeps the model, effort, web search, tools and skills it was saved with, so schedules and triggers run what you tested. On an agent with a live version, update_agent answers new_version, not live: test it with run_frozen, then set_live_version. Use: `update_agent`, `run_frozen`, `set_live_version`, `get_compilation`. [Docs](https://docs.flymy.ai/agents/guides/serverless-agents)
  Try: "Move my scheduled agent to a cheaper model, test the new version, then make it live."
- **agents** - A lead starts the agents its instruction names by id with no subagent_limits.agents list; a list you give stays the whole list. There is still no count limit. Use: `create_agent`, `run_agent`, `list_subagents`. [Docs](https://docs.flymy.ai/agents/guides/serverless-agents)
  Try: "Create a lead that hands each part of my report to the helper agents I name."
- **templates** - PromptLinks 1.3.2 shows each existing article's Google rank again: Google's results now come as redirect links, and the research agent matches them to their pages without following them. Use: `projects_templates`, `projects_copy`. [Docs](https://flymy.ai/templates/promptlinks/)
  Try: "Copy PromptLinks and find where my site can earn links."

## 1.33.1 - 2026-11-08 - Instagram Trend Watcher reports in your language

- **templates** - Instagram Trend Watcher 1.3 writes its sheet tabs, Telegram reports and Gemini notes in English, or in Russian when you write in Russian. Point it at a sheet of your own and the tabs you already use there stay as they are. Use: `projects_templates`, `projects_copy`. [Docs](https://flymy.ai/templates/instagram-trend-watcher/)
  Try: "Copy the Instagram Trend Watcher template and set it up for natgeo and mkbhd."

## 1.33.0 - 2026-11-07 - Agents v1: a named engine and helpers without a count

- **agents** - Every agent, frozen version and run names its engine, flymy/agent@1: the runtime as it is today. A frozen version keeps the engine it was frozen on and a copy keeps its source's, so a new engine can ship next to it without moving what runs. Use: `get_agent`, `list_agent_versions`, `get_run`. [Docs](https://docs.flymy.ai/agents/guides/serverless-agents)
  Try: "Which engine do my agent and its live version run on?"
- **agents** - A lead can start any of your agents as helpers, with no count limit by default. Each helper's result lists the tools it called, list_subagents shows the tools every agent always has, and a brief over 16 KB is refused with how to pass large inputs by reference. Use: `run_agent`, `list_agents`, `get_team_run`. [Docs](https://docs.flymy.ai/agents/guides/serverless-agents)
  Try: "Run my lead with five helpers at once and show which tools each one called."
- **templates** - Instagram Trend Watcher 1.2 takes the demo list on «go» and sets up its own two helpers on GPT-6 Luna. SEO / GEO 1.2.1 reads the real site of every Google result again; Website QA 1.2.1 saves every screenshot, so its visual check reads all pages. Use: `projects_templates`, `projects_copy`. [Docs](https://flymy.ai/agents/templates/)
  Try: "Copy the SEO / GEO template and check my site for three queries."

## 1.32.5 - 2026-11-06 - Cost options when you freeze an agent

- **agents** - Before you freeze, get_execution_price with cost_options shows what each run of the frozen version will cost, cheaper models of the same family priced on the same tokens with the update_agent call for each, and what made the run expensive. freeze_agent answers the same. Use: `get_execution_price`, `freeze_agent`, `update_agent`, `run_frozen`. [Docs](https://docs.flymy.ai/agents/guides/serverless-agents#before-you-freeze---check-what-each-run-costs)
  Try: "Show me the cost options of my agent's last run before we freeze it."

110 older entries: https://flymy.ai/changelog.json