How nirmalkumar31/agentic-research-engine works
Agentic Research Engine turns a question into a bounded, evidence-backed report. Its principal workflow builds an answer contract, plans and searches across question dimensions, retrieves and extracts source quotes, assesses coverage with budgeted follow-ups, then verifies atomic claims for support and relevance before publication. The CLI and web interface expose runs and reports; the web app also supports deterministic recordings. Evaluation is a separate capability. Several implementation files were not sampled, so documented workflow links are cited to the README; uncertain internal calls are omitted.
Main entry points: workflow.py Research graph · runner.py Run orchestration · research.py Search and extraction · reporting.py Synthesis and report
Components 23
User interfaces
- CLI
src/agentic_research/cli/main.py - Web API
src/agentic_research/web/api.py - Web interface
web/src/App.tsx - Published report
src/agentic_research/report.py
Research workflow
- Run orchestration
src/agentic_research/runner.py - Research graph
src/agentic_research/graph/workflow.py - Run state
src/agentic_research/graph/state.py - Answer contract
src/agentic_research/answer_contract.py - Question planning
src/agentic_research/graph/nodes/planning.py - Search and extraction
src/agentic_research/graph/nodes/research.py - Coverage assessment
src/agentic_research/answer_coverage.py - Synthesis and report
src/agentic_research/graph/nodes/reporting.py
Evidence and verification
- Source retrieval
src/agentic_research/retrieval/fetcher.py - Evidence store
src/agentic_research/evidence/store.py - Citation verification
src/agentic_research/citations/verifier.py - Relevance checks
src/agentic_research/citations/relevance.py
Runtime services
- Search providers
src/agentic_research/search/service.py - Model routing
src/agentic_research/llm/router.py - Configuration and budgets
src/agentic_research/config.py - Search service
src/agentic_research/search/service.py
Replay and evaluation
- Recorded runs
src/agentic_research/web/recordings.py - Evaluation
src/agentic_research/evaluation/ab.py
Other components
- Researcher
Connections 27
- Researcher → CLI: runs question
README.md - CLI → Run orchestration: starts run
src/agentic_research/cli/main.py - Researcher → Web interface: asks question
README.md - Web interface → Web API: streams run
web/src/App.tsx - Web API → Run orchestration: starts live run
src/agentic_research/web/api.py - Web interface → Web API: requests recordings
web/src/App.tsx - Web API → Recorded runs: loads replay
src/agentic_research/web/api.py - Run orchestration → Research graph: executes graph
src/agentic_research/runner.py - Research graph → Run state: updates state
- Research graph → Question planning: plans question
README.md - Question planning → Answer contract: builds contract
- Research graph → Search and extraction: dispatches research
README.md - Search and extraction → Search providers: runs queries
src/agentic_research/search/service.py - Search and extraction → Source retrieval: fetches sources
README.md - Search and extraction → Evidence store: registers evidence
src/agentic_research/graph/nodes/research.py - Research graph → Coverage assessment: assesses gaps
README.md - Coverage assessment → Research graph: routes follow-ups
README.md - Research graph → Synthesis and report: synthesizes claims
README.md - Synthesis and report → Citation verification: checks support
README.md - Synthesis and report → Relevance checks: checks relevance
README.md - Synthesis and report → Answer contract: judges answer fit
- Research graph → Model routing: requests model roles
src/agentic_research/llm/router.py - Run orchestration → Published report: renders result
src/agentic_research/runner.py - Run orchestration → Configuration and budgets: uses run settings
src/agentic_research/runner.py - Evaluation → Run orchestration: runs comparisons
src/agentic_research/evaluation/ab.py - Evaluation → Search providers: configures search
src/agentic_research/evaluation/ab.py - Evaluation → Model routing: configures models
src/agentic_research/evaluation/ab.py