lyx0902/apexlogic

How lyx0902/apexlogic works

ApexLogic is a resumable multi-agent deep-research system. Its main workflow takes a topic through retrieval, evidence selection, report drafting, and review; review routes the run back to research or writing until accepted, limited, or capped. Separate capabilities include persistent run recovery, source-evidence memory, PostgreSQL background execution, and report follow-up operations. The README documents several subsystem links whose implementations were not sampled, so the graph includes those documented capabilities without asserting uncertain internal calls.

Main entry points: runner.py Run orchestrator · researchers.py Research agent · storage.py Backend selection · background.py Background worker

Components 34

User entry

Streamlit UI app.py
CLI main.py

Research workflow

Research agent agents/researchers.py
Writer agent agents/writer.py
Reviewer agent agents/reviewer.py
Query planning optim/query_planner.py
Search tools tools/search_tool.py
arXiv search tools/arxiv_tool.py
Concept expansion optim/graph_expand.py
Multi-hop search optim/iterative_retrieval.py
Adaptive allocation optim/mab_search.py

Evidence and memory

Vector retrieval bge/retriever.py
Evidence reranking bge/reranker.py
Embedding client bge/embeddings.py
Memory service memory/service.py
Memory repository memory/repository.py

Execution and storage

Run orchestrator core/runner.py
Background worker core/background.py
Backend selection core/storage.py
Run repository core/run_repository.py
Postgres run store core/postgres_repository.py
Checkpoint persistence core/persistence.py
Postgres access core/postgres.py
Request cache core/cache.py

Report operations

Intent routing core/intent.py
Follow-up answers core/followup.py
Report operations core/report_operations.py
Conversation queue core/conversation.py

Other components

Researcher
LLM service
Search providers
BGE API
Redis
Research report

Connections 48

Researcher → Streamlit UI: submits topic
Researcher → CLI: submits topic
Streamlit UI → Run orchestrator: creates or resumes
CLI → Run orchestrator: creates or resumes
Run orchestrator → Research agent: runs research
Run orchestrator → Writer agent: runs drafting
Run orchestrator → Reviewer agent: runs review
Reviewer agent → Run orchestrator: routes iteration
Research agent → Query planning: decomposes queries
Research agent → Search tools: requests search
Research agent → arXiv search: requests search
Research agent → Concept expansion: expands concepts
Research agent → Multi-hop search: retrieves gaps
Research agent → Adaptive allocation: allocates results
Search tools → Search providers: queries providers
arXiv search → Search providers: queries arXiv
Research agent → Vector retrieval: selects evidence
Vector retrieval → Embedding client: embeds texts
Vector retrieval → Evidence reranking: ranks candidates
Embedding client → BGE API: requests vectors
Research agent → Memory service: recalls evidence
Run orchestrator → Memory service: publishes accepted sources
Memory service → Memory repository: reads and writes
Memory service → Embedding client: embeds evidence
Writer agent → LLM service: generates draft
Reviewer agent → LLM service: evaluates draft
Streamlit UI → Intent routing: classifies operation
Intent routing → LLM service: requests fallback
Streamlit UI → Conversation queue: queues operation
Background worker → Conversation queue: processes turns
Background worker → Run orchestrator: executes research
Background worker → Follow-up answers: answers follow-up
Background worker → Report operations: updates or verifies
Follow-up answers → Research report: returns answer
Report operations → Search providers: fetches new sources
Report operations → Research report: creates version
Run orchestrator → Research report: produces report
Run orchestrator → Backend selection: selects persistence
Backend selection → Run repository: selects SQLite
Backend selection → Postgres run store: selects PostgreSQL
Backend selection → Checkpoint persistence: opens SQLite checkpoint
Backend selection → Postgres access: opens PostgreSQL checkpoint
Run orchestrator → Checkpoint persistence: saves run state
Run orchestrator → Run repository: records run
Run orchestrator → Postgres run store: records run
Search tools → Request cache: uses optional cache
Embedding client → Request cache: uses optional cache
Request cache → Redis: reads and writes

More Python diagrams: li-chongyi/zero-dce_extension · megvii-research/motrv2 · netmanaiops/donut · pytest-dev/pytest-rerunfailures · reacher-z/awesome-rea

Drawn by GitDiagram on September 28, 2026 from the repository's file tree, README and source files. Read as Markdown ·