Edge0-AI/Edge0

Review signals: 1 connection review, 61 known vulnerability.

Reviewed as expected for a test run: resolver address lookups that send no data.

Repository processes attempted these non-loopback connections; proxies can hide the final hostname. 14 distinct locations. 8 attributed processes.

OSV reports a known vulnerability in this dependency version. 19 distinct locations. 55 findings.

OSV scanned supported lockfiles; unpinned dependencies and unsupported formats are outside this check.

A short run cannot establish that code is safe.

Connections show IP addresses; proxies can hide the final hostname.

Tracing covers dispatched commands and their children, not the whole OS; guest root can interfere with tracing.

Full safety evidence →

• One-shot chat with the edge0 CLI answers a prompt: Real input produced the expected result: Streaming inference

  Source: edge0 chat edge0-35b --prompt "Explain streaming inference in one sentence."

  Input: edge0-8b tier (edge0-35b not used: 23 GB checkpoint, 3 GB RAM guest) with the same prompt

  Expected: A one-sentence answer about streaming inference is printed

  Command: <guest-path> chat edge0-8b --prompt 'Explain streaming inference in one sentence.' --max-new 48

  Observed: Streaming inference is the technique of processing and generating responses from an AI model in real-time as data is being streamed, rather than waiting for a complete input to be processed before output begins. # 37 tokens in 92.297s

Examples

• Repository unit tests run offline (no real weights): command completed successfully

  Source: python/README.md

  Input: pytest -q in python/ venv

  Expected: all collected tests pass

  Command: <guest-path> -q

  Observed: ........................................................................ [ 39%] .......................................................ss..s............ [ 78%] ....................................... [100%] 180 passed, 3 skipped, 4 deselected in 16.15s

Local or hosted: both

Run a one-shot generation with the edge0 CLI: `edge0 chat edge0-8b --prompt "Explain streaming inference in one sentence."`, using the downloaded edge0-8b checkpoint, on Linux x86_64 CPU.

cd <guest-path> && pwd && ls -la && git status 2>&1 | head -3; python3 --version; id -u; cat /etc/os-release | head -2; ls /inputs 2>&1
cd <guest-path> && find . -maxdepth 2 -not -path './.git*' | head -100; echo ---; cat README.md 2>/dev/null | head -200
cd <guest-path> && sed -n 200,400p README.md
cd <guest-path> && git init -q 2>&1; git remote add origin https://github.com/Edge0-AI/Edge0 2>&1; timeout 120 git fetch --depth 1 origin e9cc00cfb44b0fd2abf7195dcbe707b75ea6c3e5 2>&1 | tail -5; git reset --mixed FETCH_HEAD 2>&1 | tail -3; git log --oneline -1 2>&1; git status --short | head -5
uname -m; nproc; free -g | head -2; cd <guest-path> && grep -n -A12 'requires-python\|dependencies' pyproject.toml | head -60; echo ---; (python3.12 --version 2>&1); python3 -m venv --help >/dev/null 2>&1 && echo venv_ok || echo venv_missing
cd <guest-path> && python3.12 -m venv .venv && timeout 900 .venv/bin/pip install -e '.[dev,fetch]' > <guest-path> 2>&1; echo rc=$?; tail -25 <guest-path>
cd <guest-path> && rm -rf .venv && (sudo -n apt-get update -qq 2>&1 | tail -2; sudo -n DEBIAN_FRONTEND=noninteractive apt-get install -y -qq python3.12-venv 2>&1 | tail -3); echo apt_rc=$?; python3.12 -m venv .venv && echo VENV_OK
cd <guest-path> && timeout 1500 .venv/bin/pip install -e '.[dev,fetch]' > <guest-path> 2>&1; echo rc=$?; tail -30 <guest-path>
cd <guest-path> && .venv/bin/edge0 --help 2>&1 | head -40; echo ---; ls examples scripts tests; echo ---; timeout 600 .venv/bin/pytest -q 2>&1 | tail -15; echo pytest_rc=$?
cd <guest-path> && .venv/bin/pip show mlx | head -12; .venv/bin/pip show -f mlx | grep -i 'libmlx\|\.so' | head; .venv/bin/pip index versions mlx-cpu 2>&1 | head -3; echo; .venv/bin/pip download --no-deps -d <guest-path> mlx-cpu==0.30.6 2>&1 | tail -2
cd <guest-path> && .venv/bin/pip install 'mlx-cpu==0.30.6' > <guest-path> 2>&1; echo rc=$?; tail -2 <guest-path>; .venv/bin/edge0 --help 2>&1 | head -40
cd <guest-path> && .venv/bin/edge0 models 2>&1 | head -20; echo ---; timeout 120 .venv/bin/edge0 chat edge0-8b --prompt "Explain streaming inference in one sentence." 2>&1 | tail -8; echo chat_rc=$?; echo ---; timeout 20 curl -sS -o /dev/null -w '%{http_code}\n' https://huggingface.co/Edge0/Edge0-8B-A1B-preview/resolve/main/config.json 2>&1; env | grep -i proxy | sed 's/=.*@/=***@/' | head -3
df -h /tmp | tail -1; cd <guest-path> && sed -n 1,60p scripts/fetch_models.py | grep -n 'add_argument\|def \|HF_\|token' | head -30
cd <guest-path> && nohup .venv/bin/python scripts/fetch_models.py --tier edge0-8b --target-dir models > <guest-path> 2>&1 &
sleep 60; tail -c 600 <guest-path>; echo; du -sh <guest-path> 2>/dev/null
cd <guest-path> && ls models/edge0-8b; export EDGE0_8B_MODEL=$PWD/models/edge0-8b; time (timeout 900 .venv/bin/edge0 chat edge0-8b --prompt "Explain streaming inference in one sentence." --max-new 48 > <guest-path> 2><guest-path>); echo rc=$?; tail -c 1500 <guest-path>; echo ---ERR; tail -c 1500 <guest-path>
sudo -n DEBIAN_FRONTEND=noninteractive apt-get install -y -qq g++ > <guest-path> 2>&1; echo apt_rc=$?; which g++; cd <guest-path> && export EDGE0_8B_MODEL=$PWD/models/edge0-8b; time (timeout 1200 .venv/bin/edge0 chat edge0-8b --prompt "Explain streaming inference in one sentence." --max-new 48 > <guest-path> 2><guest-path>); echo rc=$?; tail -c 1200 <guest-path>; echo ---ERR; tail -c 800 <guest-path>
grep -m3 -n 'error' <guest-path> | cut -c1-400; g++ --version | head -1; cd <guest-path> && grep -rn 'disable_compile\|MLX_DISABLE\|mx.compile' src | head -5
mkdir -p <guest-path> && printf '#!/bin/sh\nexec /usr/bin/g++ -fpermissive "$@"\n' > <guest-path> && chmod +x <guest-path>; cd <guest-path> && export PATH=<guest-path> EDGE0_8B_MODEL=$PWD/models/edge0-8b; time (timeout 1200 .venv/bin/edge0 chat edge0-8b --prompt "Explain streaming inference in one sentence." --max-new 48 > <guest-path> 2><guest-path>); echo rc=$?; tail -c 1500 <guest-path>; echo ---ERR; grep -m3 'rror' <guest-path> | cut -c1-300; tail -c 300 <guest-path>
cd <guest-path> && export PATH=<guest-path>; timeout 1200 .venv/bin/pytest -q 2>&1 | tail -8; echo pytest_rc=${PIPESTATUS[0]}; echo ---; grep -rniE 'cron|schedul|--once|--now|while True|webhook|interval' src/edge0/cli.py src/edge0/server/*.py 2>/dev/null | head -10; grep -n 'temperature\|do_sample\|greedy' src/edge0/cli.py | head -5
cd <guest-path> && sed -n 196,225p src/edge0/server/app.py; echo ---; grep -rn 'temperature' src/edge0/cli.py src/edge0/server/chat.py | head -6; grep -n 'add_argument' src/edge0/cli.py | head -30 | cut -c1-160
cd <guest-path> && mkdir -p .coldpass && ls -la .coldpass && cat > .coldpass/probe.json <<'EOF'
{"kind":"cli","argv":["<guest-path>","chat","edge0-8b","--prompt","Explain streaming inference in one sentence.","--max-new","48"],"cwd":"<guest-path>","env":{"PATH":"<guest-path>","EDGE0_8B_MODEL":"<guest-path>"},"expected_text":"Streaming inference"}
EOF
cat > .coldpass/claims.json <<'EOF'
[{"kind":"claim","claim":"One-shot chat with the edge0 CLI answers a prompt","source":"README.md","input":"edge0-8b tier with the prompt 'Explain streaming inference in one sentence.' (edge0-35b is not used: 23 GB checkpoint, 3 GB RAM guest)","expected":"A one-sentence answer about streaming inference is printed","probe_kind":"cli","argv":["<guest-path>","chat","edge0-8b","--prompt","Explain streaming inference in one sentence.","--max-new","48"],"env":{"PATH":"<guest-path>","EDGE0_8B_MODEL":"<guest-path>"},"expected_text":"Streaming inference","cwd":"<guest-path>"}]
EOF
echo written
cd <guest-path> && cat > .coldpass/claims.json <<'EOF'
[{"kind":"claim","claim":"One-shot chat with the edge0 CLI answers a prompt","source":"edge0 chat edge0-35b --prompt \"Explain streaming inference in one sentence.\"","input":"edge0-8b tier (edge0-35b not used: 23 GB checkpoint, 3 GB RAM guest) with the same prompt","expected":"A one-sentence answer about streaming inference is printed","probe_kind":"cli","argv":["<guest-path>","chat","edge0-8b","--prompt","Explain streaming inference in one sentence.","--max-new","48"],"env":{"PATH":"<guest-path>","EDGE0_8B_MODEL":"<guest-path>"},"expected_text":"Streaming inference","cwd":"<guest-path>"}]
EOF
cat > .coldpass/examples.json <<'EOF'
[{"kind":"example","claim":"Repository unit tests run offline (no real weights)","source":"python/README.md","input":"pytest -q in python/ venv","expected":"all collected tests pass","argv":["<guest-path>","-q"],"cwd":"<guest-path>","env":{"PATH":"<guest-path>"},"expected_text":"passed"}]
EOF
cat > .coldpass/cycles.json <<'EOF'
{"detected":false,"reason":"Entry points inspected: edge0 CLI (demo/chat/serve/models) and python/src/edge0/server/app.py. The only while-True loop (app.py:208) is a per-request streaming queue consumer, not a schedule. No cron, --once, --now, scheduler or webhook trigger exists in the product; serve is a long-running request server, run on demand.","checks":[]}
EOF
for f in claims examples cycles probe; do python3 -I -c "import json,sys;json.load(open('.coldpass/$f.json'));print('$f ok')"; done; ls -la .coldpass; ps aux | grep -c '[e]dge0 serve'
ls -la <guest-path> 2>&1; cd <guest-path> && <env-var>=<guest-path> timeout 1500 python3 -I <guest-path> --validate --replay > <guest-path> 2>&1; echo rc=$?; tail -c 3000 <guest-path>
head -c 1200 <guest-path>; echo; echo ...; grep -oiE '"(problems|errors|problem)": [^]]{0,300}' <guest-path> | head -5; grep -ci 'fail' <guest-path>
cd <guest-path> && cat > .coldpass/report.json <<'EOF'
{"outcome":"working","fault_owner":null,"operation":"Run a one-shot generation with the edge0 CLI: `edge0 chat edge0-8b --prompt \"Explain streaming inference in one sentence.\"`, using the downloaded edge0-8b checkpoint, on Linux x86_64 CPU.","operation_scope":"Linux x86_64 guest (Ubuntu 24.04, 2 vCPU, 3 GB RAM), CPU-only MLX build, edge0-8b tier only. The README documents the Python framework as macOS / Apple Silicon only, so this is an unsupported platform that works with the deviations below. edge0-35b (23 GB, needs more RAM) was not run. Unit tests (pytest) and one chat generation were replayed and passed.","steps":["git init; git remote add origin https://github.com/Edge0-AI/Edge0; git fetch --depth 1 origin e9cc00cfb44b0fd2abf7195dcbe707b75ea6c3e5; git reset --mixed FETCH_HEAD","cd python && python3.12 -m venv .venv && .venv/bin/pip install -e '.[dev,fetch]'","export EDGE0_8B_MODEL=$PWD/models/edge0-8b; .venv/bin/python scripts/fetch_models.py --tier edge0-8b --target-dir models","edge0 chat edge0-8b --prompt \"Explain streaming inference in one sentence.\" --max-new 48","pytest (unit tests, in python/ venv)"],"deviations":["python3.12 -m venv failed: ensurepip not available. Fixed with sudo apt-get install python3.12-venv, then reran the README install command unchanged.","pip install -e '.[dev,fetch]' succeeded but `edge0` failed at import: libmlx.so missing. The mlx 0.30.6 Linux wheel does not include the CPU library. Installed mlx-cpu==0.30.6 (same version, not in README/pyproject) into the venv.","First chat failed: MLX CPU JIT needs g++, which was missing. Fixed with sudo apt-get install g++.","Second chat failed: GCC 13 rejects MLX's generated CPU kernel (redeclaration of built-in type '_Float32'). Worked around by a PATH-first g++ wrapper at <guest-path> that adds -fpermissive. This is an environment-only change outside the repo; no product source was edited. The probe and claim env set PATH to use this wrapper.","Chat generation on CPU is slow: about 90 s for 37-39 tokens with edge0-8b.","Cycles: none. The product has no cron, --once, --now, scheduler or webhook trigger; the while-True loop in server/app.py is a streaming queue consumer. cycles.json is detected:false.","README tier example used edge0-35b; the claim uses edge0-8b because 35b is too large for this guest. Model checkpoint directory python/models (4.3 GB) and python/.venv are left in the checkout."],"stuck_step":null,"main_blocker":null,"plain_failure":null,"suggested_fix":"Documentation/product: state Linux support status explicitly (the README says macOS Apple Silicon only). For Linux CPU, the mlx-cpu package and a -fpermissive or newer-GCC build of the MLX CPU JIT are needed; the README should list these prerequisites (python3.12-venv, g++, mlx-cpu) and the GCC 13 incompatibility should be reported upstream to MLX."}
EOF
python3 -I -c "import json;d=json.load(open('.coldpass/report.json'));print(d['outcome'])"; ps aux | grep -E '[e]dge0|[f]etch_models|[p]ytest' | head -3; echo procs_checked
cd <guest-path> && <env-var>=<guest-path> timeout 1500 python3 -I <guest-path> --validate --replay > <guest-path> 2>&1; echo checker_rc=$?; head -c 300 <guest-path>; echo; grep -o '"status": "[A-Z]*"' <guest-path> | sort | uniq -c