Darkbloom's Apple Silicon provider, native inference runtime integrations, and marketing site live here. The provider runs models locally with MLX and can connect to the Darkbloom network or serve a local OpenAI-compatible API.
Put OpenCode (or Hermes) on this Mac's own models:
curl -fsSL https://api.darkbloom.dev/install.sh | bash
exec $SHELL -l # or open a new Terminal window, so your shell finds darkbloom
darkbloom --opencode # or --hermesIt asks how this Mac serves: Only me, Darkbloom network or
Both, me first. Press Enter for Both, me first. Every choice asks which
models to serve and serves them in the background; with Only me or Both, me
first, --opencode sets OpenCode up. Darkbloom network gives OpenCode no
address on this Mac. Then, in OpenCode, /models → Darkbloom. Two Macs: run
darkbloom cluster on each, then darkbloom --opencode on either Mac.
The quickstart walks through it, and Use your Mac yourself, first says what each choice runs, costs and earns. See also the local API for other OpenAI clients, installation, hardware requirements for model fit, and cache storage for encrypted SSD storage and daily write limits. To have an AI agent set up serving the Darkbloom network, point it at the agent setup prompt.
The provider is the plaintext inference endpoint. Network encryption does not hide prompts from that process or from the coordinator's transient request processing. See the encryption and storage trust model and provider attestation; do not infer security guarantees from a repository split.
Use sibling checkouts, never one repository nested inside the other:
Darkbloom/
d-inference/ Swift provider, local native APIs, MLX and landing
darkbloom-platform/ Coordinator, prompt sidecar, consumer console and admin
Centralized backend, console and admin code belongs exclusively in
darkbloom-platform.
Never implement, commit, push or upload new backend, console or admin code here.
Device-local Swift APIs remain provider code. Check git remote -v and the
intended diff before pushing.
Source ownership does not grant deployment authority. Infrastructure, hosting and traffic changes require specific human approval. Release/model registration remains an external API operation.
Tool versions are in mise.toml; supported commands are in Makefile:
mise install
make provider-build
make provider-test
make landing
make docs-checkProvider builds require Apple Silicon, the documented Xcode toolchain, and the recursive MLX dependencies. See build, test, navigation, and contribution rules. Public golden vectors check fixed expected behavior; they are not private cross-implementation qualification or live fleet evidence.
| Path | Ownership |
|---|---|
provider-swift/ |
CLI, on-device services, inference, security and tests |
libs/ |
Pinned native MLX dependencies |
landing/ |
Independent Next.js marketing site and its API routes |
fixtures/ |
Public, revision-bound test inputs |
scripts/ |
Provider builds, native qualification, install and publication helpers |
docs/ |
Provider/native explanations, references, operations and historical evidence |
See the documentation index. Console architecture, platform API and consumer guides are maintained in the platform documentation. Existing changelog history remains a record of work before and after the ownership split, not evidence of a new release or deployment.
See LICENSE and ACKNOWLEDGEMENTS.md. Published privacy and terms remain unchanged by source cleanup.