Bees
An AI application focused on running large language models locally. It includes built-in local memory and agent features, emphasizes that data never leaves the machine, and provides local inference through llama-server.
Timeline (last 90 days)
Bees' local model runs in llama-server and only listens on 127.0.0.1, so it cannot be accessed externally.
Every time Bees starts, it accesses app.bees.bot/api/config to query the login method configuration; the request carries no account information or token, and the target address can only be changed via the environment variable BEES_ACCOUNT_API_URL.
At startup and every 6 hours, Bees pulls bees-update.json from GitHub to check for updates; the request carries no user data, and there is currently no switch to turn it off.
Bees has a built-in small server that supports a free cloud AI option, running on every startup; 10 seconds after startup and then every 12 hours it accesses api.freellmapi.co to sync the free model list, and this can be disabled by setting CATALOG_SYNC_DISABLED=1.
On first startup, Bees downloads Python 3.12 (from GitHub), memory-related packages (from PyPI), and the embedding model all-MiniLM-L6-v2 (from Hugging Face) via uv to initialize local memory.
Bees' code comments explicitly state 'Never forward memories to a hosted model', and users' memory content does not appear in external requests.
When enabling the Qwen3 4B model, Bees downloads about 2.5 GB of files from Hugging Face and verifies them with a fixed checksum; users can also paste their own GGUF link.
Each Bees Agent has built-in search and page-reading tools with no switch, and search sends queries simultaneously to search.brave.com, html.duckduckgo.com, lite.duckduckgo.com, and bing.com, while news search goes through Google News RSS.
When planning, Ask Bees may query the MCP public registry registry.modelcontextprotocol.io and read skill packs from GitHub, and the Skills page uses the same calls.
After logging into a Bees Cloud team, Bees syncs once every 15 seconds, and the author observed about 20 long-lived connections to app.bees.bot on a Mac.
4 more timeline events
All Facts (14)
Bees' local model runs in llama-server and only listens on 127.0.0.1, so it cannot be accessed externally.
50%UnverifiedEvery time Bees starts, it accesses app.bees.bot/api/config to query the login method configuration; the request carries no account information or token, and the target address can only be changed via the environment variable BEES_ACCOUNT_API_URL.
50%UnverifiedAt startup and every 6 hours, Bees pulls bees-update.json from GitHub to check for updates; the request carries no user data, and there is currently no switch to turn it off.
50%UnverifiedBees has a built-in small server that supports a free cloud AI option, running on every startup; 10 seconds after startup and then every 12 hours it accesses api.freellmapi.co to sync the free model list, and this can be disabled by setting CATALOG_SYNC_DISABLED=1.
50%UnverifiedOn first startup, Bees downloads Python 3.12 (from GitHub), memory-related packages (from PyPI), and the embedding model all-MiniLM-L6-v2 (from Hugging Face) via uv to initialize local memory.
50%UnverifiedBees' code comments explicitly state 'Never forward memories to a hosted model', and users' memory content does not appear in external requests.
50%UnverifiedWhen enabling the Qwen3 4B model, Bees downloads about 2.5 GB of files from Hugging Face and verifies them with a fixed checksum; users can also paste their own GGUF link.
50%UnverifiedEach Bees Agent has built-in search and page-reading tools with no switch, and search sends queries simultaneously to search.brave.com, html.duckduckgo.com, lite.duckduckgo.com, and bing.com, while news search goes through Google News RSS.
50%UnverifiedWhen planning, Ask Bees may query the MCP public registry registry.modelcontextprotocol.io and read skill packs from GitHub, and the Skills page uses the same calls.
50%UnverifiedAfter logging into a Bees Cloud team, Bees syncs once every 15 seconds, and the author observed about 20 long-lived connections to app.bees.bot on a Mac.
50%UnverifiedAt startup, Bees passes in DSH_TELEMETRY_DISABLED=1 to disable built-in telemetry; even if enabled, data is uploaded only after the user sends feedback.
50%UnverifiedThe Content Security Policy (CSP) of the Bees application window allows only itself and 127.0.0.1, so it cannot load any remote fonts, scripts, or images.
50%Unverifiedall-MiniLM-L6-v2 is an open-source sentence embedding model of about 80MB, often chosen by local applications as the default option, and Bees' local memory feature relies on it to encode text into vectors for storage and retrieval.
50%UnverifiedThe Bees team committed to three fixes: FreeLLMAPI list syncing will run only when the user selects a free cloud model, the update check will provide a switch, and the help page will add an explanation of memory initialization.
50%