Integrate dev: execution backends, Academy multi-agent, CGFastMCP, HPC/Globus

.github/workflows/tests.yml

@@ -52,3 +52,33 @@ jobs:
     - name: Run tests
       run: |
         python -m pytest tests/ -v -k "not tblite"
+
+  # Exercises the distributed/HPC feature set: the Academy multi-agent module
+  # and the parsl / globus-compute execution backends. These tests use mocks and
+  # local subprocesses (no real HPC cluster, Globus endpoint, or LLM), so they
+  # run in CI. Live tests remain gated behind --run-globus-compute / --run-llm
+  # and are not enabled here.
+  test-extras:
+    needs: lint
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v5
+      with:
+        python-version: ${{ matrix.python-version }}
+
+    - name: Install dependencies (academy + execution backends)
+      run: |
+        python -m pip install --upgrade pip
+        pip install -e ".[academy,parsl,globus_compute]"
+
+    - name: Run academy + backend tests
+      run: |
+        python -m pytest tests/ -v -k "not tblite"

.gitignore

@@ -51,13 +51,14 @@ opencode.json
 chemgraph_mcp_logs/
 vllm/
 logs/
+runs/
+**/*.model
 error_log.txt
 .env
 test.csv
 nwchem/
 nwchem.nwi
 nwchem.nwo
-config.toml
 vib*.traj
 
 # Kubernetes secrets (keep secrets.yaml.template, ignore actual secrets)

config.toml

@@ -0,0 +1,115 @@
+[general]
+model = "argo:gpt-4o"
+workflow = "single_agent"
+output = "state"
+structured = true
+report = true
+thread = 1
+recursion_limit = 20
+human_supervised = false
+verbose = false
+
+[logging]
+level = "WARNING"
+file = "./chemgraph.log"
+console = true
+format = "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
+
+[features]
+enable_experimental = false
+enable_cache = true
+cache_dir = "./cache"
+cache_expiry = 24
+
+[security]
+validate_keys = true
+rate_limit = true
+max_requests_per_minute = 60
+
+[mcp]
+# url = "http://localhost:9003/mcp/"        # For streamable_http transport
+# command = ""                               # For stdio transport (full shell command)
+# server_name = "ChemGraph General Tools"    # Display name for the MCP connection
+
+[eval]
+default_profile = "standard"
+
+[api.openai]
+base_url = "https://apps.inside.anl.gov/argoapi/api/v1/resource/chat/"
+timeout = 30
+argo_user = ""
+
+[api.groq]
+base_url = "https://api.groq.com/openai/v1"
+timeout = 30
+
+[api.anthropic]
+base_url = "https://api.anthropic.com"
+timeout = 30
+
+[api.google]
+base_url = "https://generativelanguage.googleapis.com/v1beta"
+timeout = 30
+
+[api.alcf]
+base_url = "https://inference-api.alcf.anl.gov/resource_server/sophia/vllm/v1"
+timeout = 30
+
+[api.local]
+base_url = "http://localhost:11434"
+timeout = 60
+
+[chemistry.optimization]
+method = "BFGS"
+fmax = 0.05
+steps = 200
+
+[chemistry.frequencies]
+displacement = 0.01
+nprocs = 1
+
+[chemistry.calculators]
+default = "mace_mp"
+fallback = "emt"
+
+[output.files]
+directory = "./chemgraph_output"
+pattern = "{timestamp}_{query_hash}"
+formats = [ "xyz", "json", "html",]
+
+[output.visualization]
+enable_3d = true
+viewer = "py3dmol"
+dpi = 300
+
+[advanced.agent]
+custom_system_prompt = ""
+max_memory_tokens = 8000
+enable_function_calling = true
+
+[advanced.parallel]
+enable_parallel = false
+num_workers = 2
+
+[environments.development]
+model = "gpt-4o-mini"
+verbose = true
+enable_cache = false
+
+[environments.production]
+model = "gpt-4o"
+verbose = false
+enable_cache = true
+rate_limit = true
+
+[environments.testing]
+model = "gpt-4o-mini"
+verbose = true
+enable_cache = false
+
+[eval.profiles.standard]
+workflow_types = [ "single_agent",]
+judge_model = "gpt4o"
+recursion_limit = 50
+structured_output = true
+max_queries = 0

examples/academy/example-002-mace-ensemble-screening/README.md

@@ -0,0 +1,35 @@
+# Example 002: MACE Ensemble Screening
+
+This example demonstrates five persistent ChemGraph Academy logical agents
+running under MPI:
+
+```text
+coordinator-agent
+structure-agent-a
+structure-agent-b
+mace-agent
+assessment-agent
+```
+
+The coordinator delegates 20 SMILES candidates, structure agents generate XYZ
+files, the MACE agent runs an ensemble energy screen, and the assessment agent
+summarizes readiness/ranking evidence.
+
+The campaign assets are packaged under:
+
+```text
+src/chemgraph/academy/campaigns/example-002-mace-ensemble-screening/
+```
+
+Run it by campaign name:
+
+```bash
+chemgraph academy run-compute \
+  --system aurora \
+  --run-id aurora-mace-ensemble-screening-001 \
+  --campaign mace-ensemble-screening-20 \
+  --lm-user <argo-user>
+```
+
+See `notes.md` for the high-level architecture notes. The internal E2E user
+guide is intentionally not stored in this public example directory.