generated: '2026-07-20'
method: searched
source: https://github.com/t8/hypura#readme + https://github.com/t8/autolab#readme
notes: >-
  Community Labs ships two first-party command-line tools. Both are the primary
  interface to their respective projects; neither has a hosted service. Command
  surfaces below are transcribed verbatim from the published READMEs.
clis:
- name: hypura
  language: rust
  repository: https://github.com/t8/hypura
  description: Storage-tier-aware LLM inference scheduler for Apple Silicon. Places
    model tensors across GPU (Metal), RAM and NVMe tiers so GGUF models larger than
    physical memory run without OOM.
  install:
  - method: source
    command: git clone --recurse-submodules https://github.com/t8/hypura.git && cd hypura && cargo build --release
    binary: target/release/hypura
    requires: Rust 1.75+, CMake
  - method: homebrew
    status: announced
    note: README states "Homebrew tap coming soon" — not yet available
  commands:
  - command: hypura profile
    description: Profile host hardware (GPU working set, RAM, NVMe bandwidth). Runs
      once and caches the profile.
  - command: hypura run <model.gguf> --prompt "<text>"
    description: Run a single inference against a GGUF model.
  - command: hypura run <model.gguf> --interactive
    description: Interactive chat session.
  - command: hypura bench <model.gguf>
    description: Benchmark Hypura tier-aware scheduling against a naive baseline.
  - command: hypura inspect <model.gguf>
    description: Inspect the computed tensor placement plan without loading the model.
  - command: hypura serve <model.gguf>
    description: Start the local Ollama-compatible HTTP server on http://127.0.0.1:8080.
  flags:
  - flag: --prompt
    description: Prompt text for a single completion
  - flag: --interactive
    description: Interactive chat mode
  - flag: --max-tokens
    description: Cap generated tokens. README advises `--max-tokens 10` on untested models.
  key_flows:
  - name: Serve as an Ollama drop-in
    steps:
    - hypura profile
    - hypura inspect ./model.gguf
    - hypura serve ./model.gguf
    - Point any Ollama client at http://127.0.0.1:8080
- name: autolab
  language: python
  repository: https://github.com/t8/autolab
  license: Apache-2.0
  description: Autonomous research orchestration framework. Given a research goal it
    designs experiment campaigns, runs them, analyzes results and documents
    discoveries. Extends the autoresearch paradigm to multi-question research programs.
  install:
  - method: pip
    command: pip install autolab
    status: unverified
    note: >-
      The README publishes this command, but the PyPI name `autolab` currently
      resolves to an unrelated GPL-3.0 project (autolab-project, scientific
      instrument automation). Treat the install as source-only until Community
      Labs claims a distinct distribution name.
  commands:
  - command: autolab init "<research directive>"
    description: Initialize a research project from a natural-language directive.
  - command: autolab run campaigns/<name>.yaml
    description: Execute a campaign defined as a YAML parameter grid.
  - command: autolab status
    description: Report progress of the current research program.
  - command: autolab results --metric <name> --top <n>
    description: Query the SQLite results database, ranked by a metric.
  - command: autolab loop --backend anthropic --max-iterations <n>
    description: Start the autonomous research loop. Requires ANTHROPIC_API_KEY.
  environment:
  - name: ANTHROPIC_API_KEY
    description: Model backend credential required by `autolab loop`
  execution_backends:
  - local
  - ssh
  - docker
  - slurm