Talk to the loaded model. Tools (web search, Python, terminal) run locally in a sandbox.
Model & params
Set Model above to the provider's id (e.g.
z-ai/glm-5.2). Your key stays in this browser and is
sent only to your local OptiQ Lab, which proxies the request —
chat and tool calls work the same as with a local model.
Constrains the decode to valid JSON via lm-format-enforcer. Tools are off while JSON mode is on.
No filesystem or network isolation on this machine: the python and terminal
tools run as you. Install Docker or Podman to turn on the container sandbox.
Writes are confined to a scratch directory and network is blocked, but
reads are not restricted: tool code can read any file you
can. Use the container sandbox for a stronger boundary.
No model is being served
Load a model before chatting:
Server for one Mac, or
Cluster to shard a large
model across Macs over Thunderbolt. Or use a
cloud API above.
Start a conversation
Try ,
, or
.
🔬
Research plan
Model spent all tokens reasoning and didn't reach the answer.
Open Model & params above and set
Reasoning → off, or raise max_tokens.