VAF 0.1.0a35: local models can now also run on vLLM
VAF 0.1.0a35 brings vLLM on Linux for NVIDIA GPUs from Turing on and for supported AMD GPUs; llama.cpp stays the default.
Read moreVAF 0.1.0a35 brings vLLM on Linux for NVIDIA GPUs from Turing on and for supported AMD GPUs; llama.cpp stays the default.
Read moreVAF drafts replies to patient emails, outlines of doctor's letters and results lists, staff check each draft, and diagnosis stays with doctors.
With VAF, the AI agent and its language model run on a machine in your network, and from outside you reach it through a VPN, not an open port.
For live conversations, wake-up calls and the trip companion, Veyllo needs Android 10 or newer and a Veyllo account with a plan or credit.
In server mode, VAF 0.1.0a33 no longer answers API calls from outside the local network, even though the login screen still loads.
On Windows, VAF 0.1.0a32 no longer turns a workflow away as already running, and Stop reaches every running sub-agent there.
In VAF 0.1.0a31 the agent's web fetches no longer reach VAF itself, and a Nextcloud connected as localhost needs its network address.
VAF 0.1.0a30 refuses requests from other web pages; behind your own Nginx, take the forwarding headers from the guide.
If you had set a contact to no, block them after updating to VAF 0.1.0a29, or the agent answers once Inbound is open for their channel.
In 0.1.0a28 WhatsApp, Telegram, Discord and mail threads sit in one list and four tools go; 0.1.0a27 adds five interface languages.
VAF's daily spend limit stopped too late because the cached share went uncounted on Anthropic; 0.1.0a26 counts it.
A fresh install of 0.1.0a24 never gets past the splash screen; 0.1.0a25 builds the interface and frees the next update too.
In 0.1.0a23 a foreign client joins with one standard-library file; the sha256 and CA fingerprint in the invitation are what verify it.
In 0.1.0a22, vaf a2a create opens a room where a VAF agent works alongside others. By default it asks before anything on the machine changes.
In 0.1.0a21, vaf run --web no longer starts the background service: the dashboard ends with the session, and vaf tray starts the persistent one.
In 0.1.0a20 the per-account tool selection is enforced for the first time: where a selection is set, only those tools run, admins excepted.
In 0.1.0a19, devices on the network have to sign in: with network access on, any of them used to reach pages and APIs without a token.
In 0.1.0a18, VAF re-scans installed skills and takes high-risk ones out of the agent's reach until an admin deletes or restores them.
In 0.1.0a17, pip install --pre vaf installs the framework as a library; desktop and server still come from the installers.