How lessiyin/autoidea works
AutoIdea turns a research topic, optionally constrained by seed papers or ideas, into an evidence-grounded proposal through literature discovery, paper reading, claim and gap synthesis, idea generation and ranking, adversarial review, and feasibility assessment. Users initiate runs through the CLI or local web workbench; a LangGraph research agent coordinates specialized tools and agents. Workspace artifacts and completion checks preserve provenance. The README documents the workflow; sampled code confirms agent construction, tool registration, web run routing, and artifact auditing, but not every stage’s internal calls.
Main entry points: autoidea.py Research agent · artifact_writers.py Research artifacts · interactive.py CLI and REPL · runs.py Run manager
Components 25
User entry points
- CLI and REPL
autoidea/cli/interactive.py - Web workbench
autoidea/web/server.py - Run manager
autoidea/web/runs.py - Progress display
autoidea/stream/display.py
Research workflow
- Literature discovery
autoidea/tools/scholar.py - Paper reading
autoidea/tools/paper_content.py - Evidence and claims
autoidea/tools/cite.py - Gap synthesis
autoidea/tools/think.py - Idea generation
autoidea/tools/seed_ideas.py - Idea tournament
autoidea/tools/idea_tournament.py - Review and feasibility
autoidea/tools/think.py
Agent runtime
- Research agent
autoidea/autoidea.py - Agent prompts
autoidea/prompts.py - Model provider
autoidea/llm/models.py - Agent middleware
autoidea/middleware/ask_user.py - Conversation memory
autoidea/middleware/memory.py - Runtime settings
autoidea/config/settings.py
Run state and outputs
- Workspace paths
autoidea/paths.py - Research artifacts
autoidea/tools/artifact_writers.py - File-backed batches
autoidea/tools/batch_tasks.py - Artifact audit
autoidea/tools/artifact_audit.py - Stage completion
autoidea/tools/stage_gate.py - Pipeline state
autoidea/tools/pipeline_state.py - Session checkpoints
autoidea/sessions.py
Other components
- Researcher
Connections 32
- Researcher → CLI and REPL: starts run
README.md - Researcher → Web workbench: starts run
README.md - CLI and REPL → Research agent: invokes
- Web workbench → Run manager: routes run requests
autoidea/web/server.py - Run manager → Research agent: starts run
autoidea/web/runs.py - Research agent → Agent prompts: uses
autoidea/autoidea.py - Research agent → Model provider: configures
autoidea/autoidea.py - Research agent → Agent middleware: assembles
autoidea/autoidea.py - Research agent → Literature discovery: registers tools
autoidea/autoidea.py - Research agent → Paper reading: registers tools
autoidea/autoidea.py - Research agent → Evidence and claims: registers tools
autoidea/autoidea.py - Research agent → Gap synthesis: registers tools
autoidea/autoidea.py - Research agent → Idea generation: registers tools
autoidea/autoidea.py - Research agent → Idea tournament: registers tools
autoidea/autoidea.py - Research agent → Review and feasibility: delegates research
README.md - Literature discovery → Paper reading: selects papers
README.md - Paper reading → Evidence and claims: grounds claims
README.md - Evidence and claims → Gap synthesis: links claims to gaps
README.md - Gap synthesis → Idea generation: informs directions
README.md - Idea generation → Idea tournament: ranks ideas
README.md - Idea tournament → Review and feasibility: selects directions
README.md - Review and feasibility → Research artifacts: produces proposal
README.md - Research agent → Research artifacts: uses validated writers
autoidea/autoidea.py - Research agent → File-backed batches: uses batches
README.md - Research artifacts → Artifact audit: audits workspace
autoidea/tools/artifact_audit.py - Artifact audit → Stage completion: checks artifacts
autoidea/tools/stage_gate.py - Artifact audit → Pipeline state: validates stages
autoidea/tools/pipeline_state.py - Run manager → Session checkpoints: persists conversations
- CLI and REPL → Progress display: renders output
autoidea/cli/interactive.py - Research agent → Runtime settings: reads configuration
- Research agent → Conversation memory: uses memory
- Research artifacts → Workspace paths: writes workspace files
autoidea/tools/artifact_writers.py