Changelog

Every agent gets a preview - one or two sentences on what it does - that FlyMyAI writes each time a version goes live. People see it where they copy the agent: on its public page and, for a lead, as its project's description.

Useget_agentprojects_getDocs →

Write your own preview in the agent's Share window or with update_agent (preview, up to 280 characters); it stays through later freezes until you clear it.

Useupdate_agentDocs →

Clone an official agent template in one click. The copy brings the lead agent with its guide, scripts and a demo input - never the author's chats, helpers, data or connections - and opens a welcome chat that asks everything once, then sets up and schedules the helpers.

Useprojects_copycopy_agentDocs →

Template pages show the project diagram, every agent's tools with unit prices, how the template runs on a schedule and its measured setup cost.

flymy.ai/agents/templates →

Agents without a project are listed on the Projects page; the Agents page now opens Projects.

app.flymy.ai/projects →

Kimi K3.1 Pro, Moonshot's thinking model for agents and code, answers through /predict, the OpenAI-compatible API with function tools, and run_model, and reads images: $2 input and $10 output per 1M tokens. Set service_tier to priority for its faster Highspeed serving, or turn on private_mode.

Userun_modelget_pricing/api/v1/flymyai/kimi-k3_1-pro/openai/v1/chat/completions/api/v1/flymyai/kimi-k3_1-pro/predictDocs →

Pick Kimi K3.1 Pro as an agent's brain: in the model picker, as model kimi-k3.1-pro on the REST API, or in create_agent and update_agent on the Agents MCP. Its reasoning goes from low to max, and each model call costs $2 input and $10 output per 1M tokens.

Usecreate_agentupdate_agentlist_agent_modelsDocs →

run_model takes an optional operation_key for image and video models too. The same key and input return the first job without a second charge; the same key with other input is refused before anything runs. Text models refuse a key.

Userun_modelDocs →

POST /predict/async/ accepts an Idempotency-Key header. The same key and input return the first prediction_id instead of queueing a second paid run; another input under that key gets 409 idempotency_key_reused.

Use/api/v1/{owner}/{model}/predict/async/Docs →

Model schemas describe each input's limits, such as allowed durations, file types and minimum video size, in get_model_schema and the model's openapi.json.

Useget_model_schema/api/v1/{owner}/{model}/openapi.jsonDocs →

Use the PromptLinks starter brief to research SEO and GEO link prospects, compare publishing options and review crawler access and source credibility before choosing placements.

flymy.ai/agents/templates/promptlinks →

Use the CiteAI starter brief and downloadable runbook to plan citation placements, prepare outreach and track source-backed results with review checkpoints.

flymy.ai/agents/templates/citeai →

Read the Collaborator and DataForSEO setup guides, choose the connections for your research workflow and copy a scoped setup brief before running it.

flymy.ai/mcp/dataforseo →

Start investment research from editable Akta workflows: source private companies for a thesis, check growth signals, write a first-pass memo, compare competitors, read customer reviews or find a timely reason to reach a founder.

flymy.ai/vc →

Browse Instagram trend-watching and SEO/GEO research starter templates. Each page includes a downloadable brief, required connections, inputs, outputs and usage pricing to help you create your own agent.

flymy.ai/agents/templates →

Use dedicated Instagram and Telegram guides to choose a connector, understand its access and pricing, and copy a scoped setup request into your assistant.

flymy.ai/mcp/telegram →

Read public Instagram profiles, posts, reels, comments and search results for $0.01 per billed request with the shared FlyMyAI connection, reduced from $0.02. New agent runs, direct tool calls and Team reservations use this price.

Docs →

analyze_media, make_typed_decision and run_model for the vision model or TypeSafe Jev take an optional operation_key. The same key and input return the first answer without a second charge; the same key with other input is refused before anything runs.

Useanalyze_mediamake_typed_decisionrun_modelDocs →

When an Idempotency-Key was already used, the 409 answer carries a code: idempotency_key_reused when the key belongs to another request, idempotency_key_in_flight while the same request is still running.

Use/api/v1/agents/tasks/{id}/run-loop//api/v1/agents/compilations/{id}/run-instruction/Docs →

Give an agent instructions from an uploaded file. Helpers return complete reports by reference, and agent code can stream their contents with size and SHA-256 verification.

Useupload_agent_filecreate_agentrun_agentread_run_resultDocs →

Start an agent with its first message in the same request. Owner runs can create and schedule requested helpers, with input values that can change for future scheduled runs.

Userun_agentcreate_agentDocs →

DeepSeek V4.1 Flash answers with text and images through /predict, the OpenAI-compatible API with function tools, and run_model: $0.30 input and $1.20 output per 1M tokens. Set service_tier to priority for its fastest serving, or private_mode to keep prompts and answers out of FlyMyAI storage.

Userun_modelget_pricing/api/v1/flymyai/deepseek-v4_1-flash/openai/v1/chat/completions/api/v1/flymyai/deepseek-v4_1-flash/predictDocs →

The Python SDK reads what a run used and what it costs: client.runs.prices(run_id) for one run and client.runs.list_prices() for your newest runs or the ones you name, sync and async, from flymyai 1.2.0rc16.

UseGET /api/v1/agents/executions/{id}/prices/Docs →

Every tool call and model answer in a run's cost says whether it is settled - already charged - or still the estimate the charge will use, and model answers show their reasoning tokens.

Useget_execution_pricelist_execution_pricesGET /api/v1/agents/executions/{id}/prices/

The cost breakdown in the chat header shows reasoning tokens next to tokens out and says how many lines are not charged yet.

app.flymy.ai →

Choose DeepSeek V4.1 Flash 1017, DeepSeek's October 17 update of V4.1 Flash, in the model picker or from the Agents MCP. It takes the place of V4.1 Flash among the low-cost models for helper agents and subagents; agents already set to V4.1 Flash keep it.

Uselist_agent_modelscreate_agentupdate_agentrun_agent

An agent's live version can start from a signed webhook. Create one in the agent's Run it tab (On an event), the API or the Agents MCP, sign each JSON event with your secret, and each event ID starts at most one run. Pause it, rotate the secret or read recent deliveries at any time.

Usetriggers_v1_createtriggers_v1_listtriggers_v1_eventstriggers_v1_pause/api/v1/agents/triggers/Docs →

The Python SDK manages webhooks with client.triggers_v1: create, update, pause, activate, delete and read recent deliveries, with an idempotency key on every write.

Useclient.triggers_v1Docs →

A project shares the way an agent does: Share, then Make public & copy link gives https://app.flymy.ai/projects/<id>. Anyone with the link sees its agents, pages and what a copy needs to connect - never its data, conversations, money or connections - and Make private stops new copies.

Useprojects_shareprojects_getDocs →

Copy project brings a public project into your account: its agents with all their subagents, its pages and config, under your own budget. A tool you have not connected is asked for first; then the setup assistant connects what is left - a bot token, inputs, a schedule - in its chat.

Useprojects_copyDocs →

A run you start through the API, the SDK or the Agents MCP can create and edit your agents with create_subagent and update_subagent, as your app chat does, and start them with delegate_agent. Scheduled and frozen runs, subagents and a deployed agent's customers cannot.

Userun_agentcreate_subagentupdate_subagentdelegate_agentDocs →

telegram_bot get_me names your bot (its @username) and the chat a message without chat_id goes to. A send to a bot nobody has written to yet says to open the bot in Telegram and press Start.

Usetelegram_bot.get_meDocs →

A failed tool call in a run's steps says why, in one line (reason), for the run's owner.

Useget_runDocs →

list_agent_models names the cache-write price per 1M tokens for models that charge it, next to input, cached input and output.

Uselist_agent_modelsDocs →

Run now and Try it once on a saved version ask for its required inputs, prefilled from its schedule or its examples.

Docs →

execute_code, execute_code_with_outputs and run_code_block no longer stop at a default time limit: code runs until it finishes, its run ends, its budget or balance runs out, or the sandbox session ends (up to 30 minutes). Pass timeout_seconds to stop it sooner.

Useexecute_codeexecute_code_with_outputsrun_code_blockDocs →

Change an artifact without sending whole files: publish_artifact_version takes edits (exact text replacements) and files[].from (copy a file from an artifact you can read), with an optional sha256 check. read_artifact_files finds text, reads line ranges and describes images.

Usepublish_artifact_versionread_artifact_filesDocs →

A project's budget now also counts and caps what the project's own agent and the helpers it starts spend, as it does for the agents its flymy.yaml declares. When a cap is reached, the next paid call is refused with the usual budget message.

In your own chat an agent can test an edited code block before pinning it: run_code_block with version "draft" runs the current draft, then freeze_code_block pins it. Frozen versions keep running their pinned code.

Freeze now decides what becomes code: loops, data shaping and tool calls with clear rules turn into Python code blocks the frozen agent runs, while steps that need judgment stay with the agent. You just freeze; the saved instruction lists each code block with its version and SHA-256.

Usefreeze_agentget_compilationrun_frozenDocs →

Frozen code does not drift: editing a code block saves a new version, and every frozen version keeps running the exact code it was frozen with. The code block view in the app shows each version and which version every run used.

Userun_frozensave_instruction_versionset_live_version

Change a frozen agent without losing what works: save_instruction_version saves your edited instruction as a new version and keeps the old one, run_frozen tests it, and set_live_version makes it live. Runs and schedules that name no version use the live one; set an older one live to roll back.

Usesave_instruction_versionset_live_versionrun_frozenlist_compilations

Python SDK 1.2.0rc13 (pip install 'flymyai>=1.2.0rc13'): client.compilations.save_instruction, set_live, get_live and run_live. Over REST the same routes, and is_live=true on the compilation list finds the live version.

Use/api/v1/agents/compilations/save-instruction/{execution_id}//api/v1/agents/compilations/{id}/set-live//api/v1/agents/compilations/live/{agent_id}/run-instruction/Docs →

update_compilation pauses and resumes a version's schedule with schedule_paused, keeping its cron, labels a version with name and note, and narrows or revokes its consent for browser screenshots and files.

Useupdate_compilationget_compilation

upload_agent_file attaches a small file - a script, a prompt, a CSV - to an agent in one call, and the same operation_key returns the same file. The agent reads it in its sandbox or by name in its prompt.

Useupload_agent_filelist_agent_files

A code step can mount its recent tool results as files: run_tool_responses=true on execute_code or execute_code_with_outputs supplies the saved results within the sandbox limits, and a manifest names the tool behind each file.

Useexecute_codeexecute_code_with_outputsDocs →

Google results from DataForSEO can include the AI Overview with the pages it cites: set load_async_ai_overview on dataforseo_serp_google_organic_live.

Usedataforseo_serp_google_organic_liveDocs →

DataForSEO on FlyMyAI's account costs what DataForSEO charges for each call plus 15%. With your own DataForSEO key you pay only compute time.

Docs →

A frozen agent keeps your consent for the browser screenshots and files its source run saved, so its runs and its schedule finish without asking. The app shows that consent on the version, and Ask every run takes it back.

Usefreeze_agentrun_frozenget_compilation

Upload a file to an agent with one JSON request - filename with content or content_base64, up to 1 MiB - on the route that takes multipart uploads; the same Idempotency-Key returns the same file.

Use/api/v1/agents/tasks/{uuid}/files/upload/

Seedream 5.0 Pro creates or edits an image from a prompt and references. Its API accepts enable_safety_checker, enabled by default; explicit false selects the alternate image route. Check get_pricing for the price of each setting.

Useget_model_schemaget_pricingrun_modelDocs →

Choose Grok 4.7 as an agent brain in the model picker or from the Agents MCP. Set its reasoning level, let it call tools and keep working after context compression. Each brain call uses your FlyMy balance.

Uselist_agent_modelscreate_agentrun_agentDocs →

Grok 4.7 supports function tools and tool-result messages through its OpenAI-compatible API. Send a stable prompt_cache_key for a continuing chat.

Useget_model_schemaDocs →

Change an artifact by pasting one line into Claude: copy "<link> add these things to the artifact: " from its page and type the change. If you may edit it, you get a new version next to the live one and both stay; end with "and update current version" to make it live. Others get their own copy.

Useget_artifactread_artifact_filespublish_artifact_versionclone_artifactDocs →

Start an artifact from a standard type: a presentation, a 3D game, a landing page, a report, a dashboard or a gallery. Each works as it is and keeps its content in one file that its README.md explains, so most changes touch only that file.

Uselist_artifact_typescreate_artifact/api/v1/artifacts/types/Docs →

Save a new version next to the live one without replacing it, open it by its page link (?v= and the version number) and make it live when you are ready. get_artifact takes the artifact's own page link as well as a share link, and says what you may do next.

Usepublish_artifact_versionupdate_artifactget_artifactDocs →

Files that look like secrets stay out of your pages: env files, version control folders, keys and certificates, and text files holding another service's token are refused with the file's name, because everyone who opens a page can read its files.

Usecreate_artifactpublish_artifact_versionDocs →

Python SDK 1.2.0rc12 (pip install 'flymyai>=1.2.0rc12'): client.artifacts.types() lists the standard types, create(type=...) starts from one, and a share link with a query string or the artifact's own page link works wherever the SDK takes a link or an id.

Use/api/v1/artifacts/types//api/v1/artifacts/Docs →

In your FlyMyAI app chat, your agent can create or edit your agents using your connected tools, then run them as subagents. New chats use a shared $5 subagent budget by default; you can change it before running.

Docs →

Seedance 2.5 by ByteDance makes videos with sound of 4 to 30 seconds at up to 1080p from a prompt, a first and last frame, or up to 30 reference images, 10 clips and 10 audio tracks; frames and references may show real people. A 5-second 720p clip costs about $1.22.

Userecommend_modelget_pricingrun_model/api/v1/flymyai/bytedance-seedance-2_5-reference-to-video/predict/async/Docs →

YoTrends finds what is trending on YouTube and TikTok: viral videos by platform and region, the creators that trend most, and content categories with their trend weight and trendlines. It runs on FlyMyAI's YoTrends plan with nothing to set up; a call costs compute time.

Usesearch_toolsexecute_toolDocs →

Gemini 3.1 Flash Lite, Gemini 2.5 Flash, Gemini 3 Flash, Gemini 3.1 Pro and Grok 4.1 Fast are in the model catalog as chat models, priced per 1M tokens: recommend_model suggests them, run_model runs them, and agents in the app can be given them in the model picker.

Userecommend_modelflymyai_modelsget_pricingrun_model

run_model answers with text models that only stream, such as Gemini and Grok 4.1: one paid call returns the whole answer and its token usage.

Userun_model

instagram_post_info lists every slide of an album post in order, with each image or video URL, ready to pass to analyze_media.

Useinstagram_post_infoanalyze_mediaDocs →

Nano Banana 2.1 by Google makes and edits images up to 4K from a prompt and up to 14 reference images, with readable text and characters that keep their look across edits. You pay Google's token price: $0.0336 per 1K image plus thinking tokens, about $0.04 a run.

Userecommend_modelget_pricingrun_model/api/v1/flymyai/nano-banana-2_1/predictDocs →

A lead agent sees and reads the files its subagents save and can pass them straight into its code run: the code finds them under /tmp/workspace/inputs, so collected data is processed without copying it by hand.

Userun_agentget_runflymy.ai/skill.md →

Instagram data in Agent Teams on FlyMyAI's shared account costs the same as in a single agent: $0.02 per request, what the data provider charges. With your own HikerAPI key you pay only compute time.

Userun_agentDocs →

A subagent that reaches its budget or keeps repeating the same call is stopped with its saved result and the reason, and its lead finishes with what the subagents delivered instead of redoing their work or waiting.

Userun_agentget_runget_team_runflymy.ai/skill.md →

An API, frozen or scheduled run that cannot do part of its task, such as reading a file that does not exist, finishes as completed and names the missing part instead of failing.

Userun_agentrun_frozenschedule_agentflymy.ai/skill.md →

Choose the model of a project's agent: the Model list in the project panel, `model` in projects_agent, the REST endpoint or the Python SDK 1.2.0rc11. A new project agent starts on the strongest model, and your choice stays through platform updates.

Useprojects_agentlist_agent_models/api/v1/agents/projects/{id}/agent/Docs →

Instagram data on FlyMyAI's shared account costs what the data provider charges for each request: $0.02 per request, and the balance check is free. With your own HikerAPI key you pay only compute time.

Usesearch_toolsexecute_toolDocs →

A project's spend now includes its project agent: the model and tool calls of its chats and of the subagents it starts count next to page calls and servers.

Useprojects_getprojects_listDocs →

When a vision model refuses a request, the answer says why in a reason field next to the usual error, for example a prompt longer than 100,000 characters.

Useanalyze_media/api/v1/flymyai/gemini-3_8-flash-vlm/predictDocs →

A shared artifact link now opens the page under one thin bar. With sources shared, Make your own version gives copy-ready steps for Claude Code, Codex and chat apps: connect FlyMy.AI, paste one prompt, and your agent clones it and publishes your version. Agents fetching the link see them too.

Useclone_artifactread_artifact_filespublish_artifact_version/api/v1/artifacts/shared/{handle}/Docs →

Python SDK 1.2.0rc9 (pip install 'flymyai==1.2.0rc9') adds client.mcp_servers: add an MCP server by URL, sign in to an OAuth server with a link a person approves, then call its tools from your code. Tokens and the client secret never come back.

Use/api/v1/agents/mcp-servers//api/v1/agents/mcp-servers/{id}/authorize/Docs →

Connect Slack once and agents work as you: search messages and files, read and post in channels and direct messages, share files, write canvases and lists, and set reminders. The bot connector is now called Slack Bot and posts as the FlyMyAI app.

Usesearch_toolsexecute_toolDocs →

Wan 3.0 by Alibaba makes video with sound from a prompt, from a first and last frame, or from reference images, clips and audio: 2-30 seconds or a length it picks, at 480P, 720P or 1080P. You pay per generated second, $0.05 to $0.20 by resolution.

Userecommend_modelget_pricingrun_model/api/v1/flymyai/wan-3_0-t2v/predict/async/Docs →

Grok 4.7 by xAI answers prompts and chats with text and images through /predict, the OpenAI-compatible API and run_model. Choose how hard it reasons, from low to xhigh, and turn on private mode to keep the prompt and the answer out of FlyMyAI storage.

Userun_modelget_pricing/api/v1/flymyai/grok-4_7/openai/v1/chat/completions/api/v1/flymyai/grok-4_7/predictDocs →

Add an MCP server that signs in with OAuth, such as Linear's or Notion's: FlyMyAI registers itself, you approve access in the provider's window, and access renews on its own. A server without automatic registration takes the client ID and secret of your own OAuth app.

Useadd_mcp_serverauthorize_mcp_serverconnect_mcp_serverDocs →

Connect Collaborator.pro with your API token. Agents read your link-building deals as an advertiser or a site owner, and search its catalog of sites for guest posts by price and SEO metrics once Collaborator opens the catalog for your account.

Usesearch_toolsexecute_toolDocs →

Instagram Public Data returns an account's photos, albums and reels in one list: pinned posts first, then newest first. Posts and reels include shares; posts also include reposts and saves when Instagram returns them.

Usesearch_toolsexecute_toolDocs →

Ask your assistant to clone an app you own. Preview its latest successful release, then create a private copy with new resources and a separate budget. Stored data, chats and personal connections stay with the original.

Useprojects_planprojects_createDocs →

Project workspaces use the full page width. Move between sections from the Projects menu, and drag the divider to adjust chat width beside the graph. Your browser remembers the width; the account menu stays on the right.

Useprojects_listDocs →

Clone your own app in its workspace. Review the plan, then create a private copy with new resources and its own budget. The copy uses the latest successful release; stored data, conversations and personal connections are not copied.

Useprojects_listprojects_getDocs →

Cloud agents save private checkpoints before compacting a conversation. Recovery pointers let them retrieve earlier decisions, evidence and unfinished work in bounded chunks; ordinary chats can find prior checkpoints of the same agent.

Docs →

Open a project with a searchable project picker, one agent list, a resource graph and a visible master chat. Select an agent to read its conversation and earlier runs without leaving the workspace.

Useprojects_listprojects_getDocs →

A project is one flymy.yaml: its pages, agents, storage and budgets. Plan it first to see what it creates and its price per hour; nothing is created until you confirm, and every change you apply is a new release.

Useprojects_planprojects_createapps_planapps_applyprojects_listDocs →

Publish a small web page, game or deck as an artifact. Every change is a new version and you pick the live one; share it by link, or invite people by email or username to view or edit it.

Usecreate_artifactpublish_artifact_versionshare_artifactadd_artifact_memberclone_artifactDocs →

An artifact someone shares with you by name waits as an invitation: nothing of it opens until you accept, and you can decline. The owner sees you only as the email or username they typed.

Uselist_artifact_invitationsaccept_artifact_invitationdecline_artifact_invitationDocs →

Publish a page as a site and share it by link or with people by email. Its declared calls to models and agents are paid by your sponsor pool or by visitors who agree to pay, within daily caps.

Usepublish_siteshare_siteinvite_to_siteget_site_usageDocs →

Direct tool calls - execute_tool, the REST tool call, and compiled and scheduled runs - are charged like agent runs. A paid call needs a positive balance; with an empty balance it is refused before it runs.

A failed scheduled run's page says why it failed and lets you change the instruction of the version that ran (the schedule stays on it), run it again or resume the schedule.

Docs →

Failure emails open the failed run itself, quote the agent's last message and say how many runs before it succeeded. A run that stops without a result reports the agent's own last words as its error.

Useget_runread_run_result

When a scheduled run fails, the agent's owner gets an email with the reason and a link to act on it: reconnect a connection, top up the balance or open the agent.

Docs →

After 3 failed scheduled runs in a row the schedule pauses itself and keeps its time. get_compilation shows schedule_paused and the reason; resume from the agent page, with schedule_paused false, or by setting the cron again.

Useget_compilationupdate_compilationPATCH /api/v1/agents/compilations/{id}/Docs →

Every sharing action works from the FlyMyAI connection in Claude, ChatGPT or Grok: see what you can share, share by email or with a teammate, accept, decline and revoke. list_shareable_mcp_connections and list_mcp_share_people help pick what and with whom.

Uselist_shareable_mcp_connectionslist_mcp_share_peopleinvite_mcp_userrevoke_mcp_accessDocs →

Instagram, Facebook, LinkedIn, Reddit, TikTok and YouTube connections can be shared like any other connection.

Useinvite_mcp_user

See every call made through a share - what ran, when and what it cost: Calls in Shared by me, Your calls in Use, or list_mcp_share_activity in chat.

Uselist_mcp_share_activityDocs →

Connections other people share with you work in Claude, ChatGPT, Grok and any MCP client signed in to FlyMyAI: list_configured_tools lists them next to your own under shared_with_me, with who shared each, and call_shared_mcp calls them. No separate token or snippet.

Uselist_configured_toolslist_shared_mcp_toolscall_shared_mcpDocs →

A connection shared with you has no monthly call cap: you pay for your own calls from your balance until the owner revokes it or the access period ends.

Usecall_shared_mcp

Your fleets are on the Agents MCP: projects_list shows each fleet - a lead agent and the agents it starts - with its status, runs, spend and failed runs today, this week and this month; projects_get opens one with its agents, schedules, tools and diagram; projects_errors reads its errors.

Useprojects_listprojects_getprojects_errors

Tool calls priced by running time - Notion, Slack, GitHub, Google, CRMs and other connectors, including calls through a connection shared with you - cost $0.000037 per second ($0.133 per hour). Search, media and voice tools keep their own price.

Useexecute_toolcall_shared_mcpflymy.ai/pricing →

Code in the sandbox and audio or video edits with media_editor cost $0.0000185 per second ($0.0666 per hour). execute_code and media edits are billed for the seconds the call takes, run_code_block only for its execution segment.

Useexecute_codeexecute_code_with_outputsrun_code_blockcreate_video_from_imagesreplace_audioflymy.ai/pricing →

GPT-5.6 Sol costs $4 input and $20 output per 1M tokens. On GPT-5.6 models, input written to the prompt cache bills at 1.25x the input rate, as on GPT-6 and Claude.

Uselist_agent_modelsflymy.ai/pricing →

Projects lists each fleet - a lead agent and the agents it starts as subagents or in its Team - with a diagram of the lead, its agents, their models, tools, schedules and budget, and what they ran and spent today, this week and this month.

UseGET /api/v1/agents/projects/GET /api/v1/agents/projects/fleet:<lead id>/app.flymy.ai/projects →

A fleet's errors journal groups the failed runs of each of its agents with the newest error and a link to that run, and the Agents tab says which fleet a chat belongs to.

UseGET /api/v1/agents/projects/fleet:<lead id>/errors/

New browser sessions use gpt-6-luna by default. Pick a newer browser model with model, for example gpt-6-sol or claude-opus-5. code_mode and use_own_llm_key keep working on gpt-5.6-luna.

Userun_browser_tasksend_taskDocs →

invite_mcp_user shares a connection with no end date: access_days 0 keeps access until you revoke it. 1 to 30 days work as before.

Useinvite_mcp_userrevoke_mcp_access

cancel_mcp_invitation also declines an invitation sent to you; its link then stops offering Accept.

Usecancel_mcp_invitation

invite_mcp_user shares a connection of any supported service by email. The recipient can use every one of its actions, reads and writes, on your account; they pay for their own calls, your credentials stay with FlyMyAI, and you can revoke it at any time.

Useinvite_mcp_userlist_shared_mcp_toolscall_shared_mcp

A recipient's own agents can use the actions of connections shared with them, reads and writes, in their runs.

Userun_agent

Team connections share every action the owner selects, reads and writes. Built-in FlyMyAI tools and tools that run on FlyMyAI's own key are not shared.

gemini-3_8-flash-vlm takes media_resolution low, medium or high. low reads an image with about a quarter of the input tokens; a video is already read at low by default.

Useanalyze_media/api/v1/flymyai/gemini-3_8-flash-vlm/predict

A public lead agent is cloned together with the agents it runs as subagents, all or none, and the copy starts your copies. The share page lists each agent with its model and tools.

Usecopy_agentset_agent_public

gemini-3_8-flash-vlm takes a prompt and a system_instruction of up to 100,000 characters each.

Useanalyze_media/api/v1/flymyai/gemini-3_8-flash-vlm/predict

whoami shows pending_usd, the usage still being charged, and available_usd, the balance after it.

Usewhoami

A run's price shows completion review, tool discovery and context compression on their own lines, and lists the models the run ran through run_model.

Useget_execution_price

run_model gives each generated file a url that opens it without signing in for up to 15 minutes, so an image made in chat can be viewed and saved right away.

Userun_model

analyze_media takes the Gemini 3.8 Flash controls: response_schema, thinking_level, temperature and seed, and fps, start_offset and end_offset for a video.

Useanalyze_media

freeze_agent and schedule_agent take guidance: plain text the compiler follows, such as which steps and inputs must stay as they are.

Usefreeze_agentschedule_agent

gemini-3_8-flash-vlm takes response_schema, a JSON Schema its answer must follow, and thinking_level low, medium or high. low spends fewer thinking tokens, which suits labeling and extraction.

Use/api/v1/flymyai/gemini-3_8-flash-vlm/predict

Its answers include finish_reason, so MAX_TOKENS shows that max_output_tokens cut the answer, and usage.thinking_tokens. temperature and seed tune sampling; for a video, fps, start_offset and end_offset choose what is read.

The Models API names every field this model does not use in warnings instead of ignoring it.

Use/api/v1/flymyai/gemini-3_8-flash-vlm/predict

A run can start your other agents as subagents with delegate_agent: each works with its own model, prompt, tools and MCP servers, and the lead collects their results. Choose which agents it may start in subagent_limits.agents or in the chat Subagents control.

Userun_agentrun_frozen

delegate_agent without an agent runs a copy of the lead with its tools on another model from list_agent_models, for example a cheaper model for simple parallel steps.

Uselist_agent_models

Each subagent gets its share of the run's subagent budget and stops when it is spent. get_team_run shows every subagent's agent, model and spend, and subagent_status returns a large result in full.

Useget_team_run

A lead can freeze and schedule its subagents with freeze_subagent and schedule_subagent. A frozen or scheduled lead starts the frozen versions of its agents that it pinned when it was frozen, so editing an agent does not change a schedule.

Usefreeze_agentschedule_agentget_compilation

Model pages and model cards show each model's price, computed from the same rules that charge the run: per run, per second of video, per image, per 1M tokens or per hour of run time, with a table when inputs like resolution or duration change it.

app.flymy.ai/models →

get_pricing quotes any listed model, and get_model_schema, flymyai_models and recommend_model show the same price: one run with default settings, the range and the inputs that change it.

Useget_pricingget_model_schemaflymyai_modelsrecommend_model

Model endpoints in the REST API return price, and chat models also return openai.base_url for the official OpenAI SDK. GET /api/v1/demo/catalog/ lists every public model with both.

Use/api/v1/demo/catalog/Docs →

Chat model pages have an OpenAI tab with ready Python, JavaScript and cURL code for the OpenAI-compatible route: keep your OpenAI code and change the base_url and key.

Docs →

read_run_result reads the whole result or error of a finished run when it is longer than the preview get_run shows, in parts of up to 64 KiB, using the ref from get_run. The sha256 from get_run confirms you have every byte.

Useget_runread_run_resultDocs →

When a direct execute_tool call is turned down before anything happens - wrong arguments or filters, missing setup, or the service refusing the request - the answer says rejected and gives the reason. Correct the call and send it again with a new operation_key.

Useexecute_toolDocs →

run_agent and run_frozen answer right away with a short receipt: the run id, status, run_seq and subagent limits. Follow the run and read its output with get_run.

Userun_agentrun_frozenget_runDocs →

list_configured_tools returns one short row per connection - id, alias, toolkit and status - 25 to a page by default. Filter by toolkit or pass view=full to get setup steps and the connect link.

Uselist_configured_toolsDocs →

Connections other people share with you appear in My MCPs next to your own, with who shared them and their email. Accept or decline an invitation right there; the owner gets an email when you accept. Pilot: Linear, Notion, Calendly and Gmail reads.

Uselist_mcp_shareslist_mcp_invitationsaccept_mcp_invitationDocs →

Share a connection for 1, 7, 14 or 30 days, or in the app with no end date until you revoke it.

Useinvite_mcp_userDocs →

Agents use your own connections first and can use connections shared with you: with no Notion of your own, a Notion request uses the shared one; ask for someone else's or a shared connection to get theirs. Calls are billed to your balance.

Docs →

One catalog of generative models: images, video, audio and music, speech, upscaling, embeddings and LLMs. recommend_model picks one for a task, get_pricing shows the price, run_model runs it on your key.

Uselist_media_modelsrecommend_modelget_pricingrun_modelDocs →

Build one agent from an idea: create it, run it and read the run, refine the same agent, then freeze an accepted run into a reusable release and run that release. A frozen agent keeps running in the cloud when your laptop is closed.

Usecreate_agentrun_agentget_runfreeze_agentget_compilationrun_frozenDocs →

Open a media template from the gallery at flymy.ai/media (product ads, unboxings, outfit reveals and more), run it as is, clone it and make it your own agent.

Docs →

Search the connector catalog by what you need done (Slack, Notion, GitHub, Google Workspace, CRMs, payments, web search, browser automation and more), read the exact schema, then run an action directly or inside an agent.

Usesearch_toolsget_tool_schemasexecute_tooladd_toolDocs →

Pilot: share one connected account with another person by email from the MCP page in the app. They use only reviewed read actions and pay for their own calls; your credentials never leave FlyMyAI. Sharing with a team is in a private pilot.

Docs →

Share an agent as a template: set_agent_public makes it public, and copy_agent forks an agent into your own account to change as you like.

Useset_agent_publiccopy_agent

Run a frozen agent on a schedule, or call it from your product over the Agents REST API or the Python SDK (package flymyai).

Useschedule_agentrun_frozenDocs →

An agent can hand parts of a run to subagents and join their results in the same run. subagent_limits on run_agent caps their spend, count and parallelism before anything starts.

Userun_agent

TypeSafe Jev 1.13 answers typed questions quickly and cheaply: pick one option, give a yes probability or a score. Use it for routing, triage and guardrail checks from an agent or directly.

Usemake_typed_decisionDocs →

Connect several accounts of one service under aliases, group connections into MCP resource sets and grant a set to agents or to an agent group.

Useadd_toolcreate_mcp_resource_setreplace_mcp_resource_set_memberscreate_agent_groupDocs →

Connect an MCP client to one server, https://mcp-agents.flymy.ai/mcp, with OAuth sign-in or an API key sent as X-API-Key. It carries models, tools and agents; whoami confirms which account is connected.

Usewhoamisearch_toolsDocs →

Bring your own MCP server: register it, connect it and call its tools from chat or from an agent.

Useadd_mcp_serverconnect_mcp_servercall_mcp_serverflymy.ai/skill.md →

Embed an agent in your product with one builder key: publish a frozen agent as a deployment, identify each of your users by external_user_id, let them connect their own accounts through a hosted link, and run and meter per user.

UsePOST /api/v1/agents/deployments/{deployment_id}/run/flymy.ai/skill.md →

Deploy your own Python model as an auto-scaling GPU endpoint with the FlyMyAI CLI (fma) and pay only for the compute it uses.

Docs →