RAG starters · Rank #4
deepdoc
Grade B: derived only from the two verdicts below. How grades work
Deep research tool for local knowledge base.
Why this verdict
How we check →- 1 commit by 1 contributor in the last 90 days (1 in the last 30). receipt ↗
- LangChain 0 declared; latest is 1 (1 major behind). receipt ↗
- Not counted in verdict: LangGraph 0 → 1. receipt ↗
Snapshot: commit 42c8758 on main, checked 2026-10-19 06:31 UTC. Receipts link to files at this exact commit.
Dependency freshness
| Package | Declared | Latest | Status |
|---|---|---|---|
| LangChain langchain | 0.3.22 | 1.4.6 | 1 major behind |
| LangGraph langgraph | 0.3.24 | 1.2.16 | Behind, not counted |
Declared = lowest version the kit's dependency file (package.json, pyproject.toml or requirements.txt) allows. Latest = the npm "latest" tag or the newest PyPI release. TypeScript, the Stripe SDK, FastAPI and other 0.x packages are shown but not counted in the verdict.
What's in the stack
Detected from declared dependencies. A listed package means it's installed, not necessarily fully wired up.
What the README says you get
- AdminConfigurable thresholds
customize model behavior, parallelism, retrieval size, reflection limits, and Jev thresholds in configuration.py
- AIReport planner
A report planner creates the initial structure from your topic and outline.
- AIPlan approval step
You can approve it or request changes before research starts.
- AIPer-section research agents
One research agent runs for every report section.
- AIJev relevance judging
Uses Jev to judge relevance, usable evidence, contradiction, and prompt-injection-like content.
- AICandidate filtering and reranking
Filters and reranks the candidates before evidence reaches the writing LLM.
- AIEvidence gap reflection
Uses Jev reflection scores to find subsections that still lack evidence.
- AICoverage retry queries
Generates new queries for missing coverage, up to the configured reflection limit.
- APIBounded retry policy
Jev requests use a bounded retry policy for temporary 429/5xx, connection, and timeout failures.
- ContentSource metadata retention
keeps page and source metadata
- ContentMarkdown report export
combines all completed sections and exports a Markdown file
- DataQdrant vector search
creates searchable chunks in Qdrant
- DeployDocker Compose services
start the required services (e.g., Qdrant) using
Extracted from the kit's README by a local AI model (deepseek-flash) on Sep 29, 2026. Every item is checked against an exact README quote; 1 unverifiable item was discarded. A README claim is the maker's description, not something we tested.
Watch deepdoc
Get told if it goes dormant, gets archived, falls further behind on a core dependency, or changes its license or price.
History
Sep 29, 2026 → Oct 19, 2026. One square per check, colored by maintenance status. Hover for details.
Show as a table
| Date | Maintenance | Dependencies | Commits (90d) |
|---|---|---|---|
| Oct 19, 2026 | Active | 1 major behind | 1 |
| Oct 12, 2026 | Active | 1 major behind | 1 |
| Sep 29, 2026 | Active | 1 major behind | 1 |
Sources
Every claim on this page comes from these public records, checked . Open them to verify.
- Commit history ↗ default branch (main) at the checked commit: pulse, contributors, last commit
- Releases ↗ latest release
- requirements.txt ↗ declared dependency versions and stack (at commit 42c8758)
- README ↗ quoted features at the checked commit
- PyPI ↗ latest versions of Python packages (each linked in the table above)