schema = "m64/capsule@1"
description = "opencode agent on Qwen3.8-Flash-Next (Basalt, RTX 5090). opencode 1.18.23 coding agent driving Qwen3.8-Flash-Next (IQ3_S pack, 384k context) on the Basalt engine, both in one container on one RTX 5090 with 64 GB of RAM. Each session gets an empty directory; the agent can write and run code there. opencode itself (its terminal interface, in the browser, opened on the folder home) is on a second port. Needs no network (the model runs in the same container); serve it with --allow-external-network to let the agent install packages."
kinds = ["agent.session"]
authors = ["dh7"]
license = "mit"

[hardware]
platform = "linux-x86_64"
gpu = "NVIDIA GeForce RTX 5090"
gpu_count = 1
vram_gb = 30
# The pack's experts (46.8 GiB) sit in system RAM next to what VRAM caches: a 64 GB
# machine (which reports ~60.5 GB).
ram_gb = 60
driver = ">=580"
disk_gb = 90

[runtime]
sandbox = "container"
# No image: the Dockerfile builds Basalt at a pinned commit on nvidia/cuda 13.0.0 (pinned by
# digest) and adds node 22 + opencode.
gpu = true
shm_gb = 8
internal_port = 9500
health = "/health"
# Basalt loads first (~30 s with the pack in the page cache, minutes cold), then opencode.
expected_load_seconds = 300
user = "1000:1000"

# opencode's terminal interface in the browser (ttyd), opened on the folder home, on a second
# front port. Raw: no per-session isolation and no params filtering there.
[runtime.app]
internal_port = 7681

# The Basalt pack is the whole model (weights, tokenizer, chat template, vision, drafter).
[[artifacts]]
type = "hf"
repo = "jesdga/Qwen3.8-Flash-Next-Basalt"
revision = "2a6a758dc12f1d027d2d35b8964110df7e80f70b"
include = ["LICENSE", "Qwen3.8-FN-GSQ-RCO-IQ3_S.basalt"]
mount = "/weights/basalt"