Conversation
Update dev from main
Add RAG Agent to ChemGraph
Add session memory persistence with CLI session management.
Add evaluation & benchmarking module with LLM-as-judge and documentation
Add OpenCode MCP configuration for using ChemGraph tools directly
Add missing evaluation documentations and update config.toml
Add structured output evaluation with checkpointing and FormatterAgent retry
Collaborator
Author
|
@keceli I will keep this PR open for now so I can track the major changes before asking you to review it. |
tdpham2
changed the title
… logging/config bugs - Move CLI from src/ui/cli.py to src/chemgraph/cli/ (main.py, commands.py, formatting.py) - Extract shared async helper into chemgraph.utils.async_utils - Update pyproject.toml entry point: ui.cli:main -> chemgraph.cli:main - Merge supported_argoproxy_models into supported_argo_models with argo: prefix - Switch Groq to prefix-based routing (groq:<model>) instead of curated list - Add ALCF model support with Globus OAuth auth flow and default base URL - Fix logging to use stderr instead of stdout (critical for MCP stdio transport) - Add configure_logging() for package-wide verbosity control (-v/-vv flags) - Replace bare print() debug statements with proper logger calls - Add --base-url CLI flag and workflow aliases (python_repl -> python_relp) - Use ThreadPoolExecutor for cross-platform timeout instead of Unix signals - Add default section merging to config loader so partial configs don't crash - Add [api.alcf] config section and DEPLOYMENT.md - Add dependency_tests.yml CI workflow
- anthropic.py: Replace all OpenAI references with Anthropic in logs, prompts, docstring, and critically fix env var from OPENAI_API_KEY to ANTHROPIC_API_KEY so auth retry actually works - openai.py: Fix argo:gpt-5.4 wire name from 'gpt52' to 'gpt54'
Refactor CLI, unify model routing, and fix logging/config bugs
… logic - Rewrite multi_agent.py with LangGraph Send() pattern for parallel executor fan-out (Planner -> Send(executor_subgraph) -> Planner) - Replace sequential worker loop with independent executor subgraphs that each run a ReAct loop (executor_agent -> ToolNode -> executor_agent) - Add retry logic with error feedback for planner and response agent JSON parsing failures - Reduce multi_agent_mcp.py to a thin wrapper delegating to construct_multi_agent_graph (ToolNode handles MCP tools natively) - Rename formatter_max_retries to max_retries for consistency - Update PlannerResponse schema, state definitions (ExecutorState, PlannerState), and multi-agent prompts for new architecture - Update single_agent.py and ase_tools.py - Update tests for new schemas and add planner fallback retry tests
…gent Since LangGraph's ToolNode handles both sync LangChain tools and async MCP tools natively, the separate multi_agent_mcp workflow is redundant. The multi_agent workflow now works identically with MCP tools passed via the tools parameter. - Delete src/chemgraph/graphs/multi_agent_mcp.py - Remove multi_agent_mcp import, workflow_map entry, and dispatch branch from llm_agent.py - Remove from CLI workflow choices and eval config valid types - Remove from test_graphs.py and test_graph_constructors.py - Update docs/evaluation.md and SKILL.md workflow tables
Replace duplicated MCP server code in example scripts with references to the canonical chemgraph.mcp.mcp_tools module. This eliminates ~960 lines of duplicated tool definitions that were drifting out of sync. - Delete scripts/mcp_example/mcp_stdio/mcp_tools_stdio.py (484 lines) - Replace scripts/mcp_example/mcp_http/start_mcp_server.py (486 lines) with a 24-line thin wrapper importing chemgraph.mcp.mcp_tools - Update stdio run scripts to spawn 'python -m chemgraph.mcp.mcp_tools' instead of referencing local server copies - Add scripts/mcp_example/mcp_stdio/run_chemgraph_multi_agent.py for local multi-agent MCP testing using the multi_agent workflow - Update start_mcp_server.sub to use the module directly - Update both README files with corrected instructions and port numbers
Refactor multi-agent to Send()-based architecture, remove multi_agent_mcp, and clean up MCP examples
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Major improvements for ChemGraph: