Eleven short tours of the current interface, recorded against a live Gateway with a prepared sample project. Follow the plan, explore eight evolvable entity types, and move between Chat, Canvas, Code and Science.
Overview brings the living plan, Runtime activity, capabilities, candidates and project files together. Search projects from the sidebar and pick up the next step in the same workspace.
Overview, Chat, Canvas, Code and Science provide different ways to work with one project. Switching views keeps the project resources together.
The Canvas is a node editor for the times a picture beats a paragraph. Components are dragged in from a categorised palette and joined into a flow you can save and reuse — a human-facing library, separate from the agent-facing workflows.
Code opens an OpenVSCode server on the project workspace. Prepare the local base image once; the first Code launch builds and starts the editor container. Later launches reuse the image and installed extensions.
Run Python on the project files and inspect the result in the Notebook panel. The REPL, JupyterLab and the agent’s code interpreter use the shared project kernel. This clip runs a small example on synthetic data.
.ipynbThe Overview cards lead to the capability library and shared session notes. Library availability, session selection and candidate staging describe different states; each remains visible in its own place.
Four starting points cover the usual asks — review, plan, explain, investigate — and the composer underneath takes anything else. Files can be attached to the session before the run begins.
Chat’s right-hand panel provides file previews, task activity and details of selected execution steps. The prepared sample is idle, so the recording shows the file viewer and the empty activity and inspector states.
The sidebar groups the Browser and Computer live-view launchers with configured remote machines and deployments. This clip shows the local launch controls; available remote hosts depend on the session configuration.
Browse registered providers and inspect each model’s advertised tool, vision and streaming support. Provider connections and model settings can be configured from the same workbench.
The theme is remembered across sessions, and the endpoint is not baked in: point the workbench at a gateway on another host, or supply a token when the gateway requires one.
AGENTEVOLVER_GATEWAY_TOKEN for protected gatewaysPrepare the base image once, then launch the Python Gateway and Vite together inside the base sandbox. For a host conda setup, follow the Frontend README.
docker build -f docker/base/Dockerfile -t agentevolver/base:latest . scripts/run-in-sandbox.sh -- scripts/serve-ui.sh
Then open http://127.0.0.1:5173. The sandbox runs with --network host, so the container loopback is the host loopback; the UI connects to ws://127.0.0.1:9876/ws by default.