Research operating system
A research workspace where method governs the AI.
Protocols, codebooks, extraction cells, analyses, and audit trails are structured objects that agents operate under your supervision.
Meta-analysis can change a field. The technical burden around it should not be the price of admission.
Meta-analysis can reveal what no single study can. Done well, it consolidates a field's empirical record, exposes inconsistencies in theory, and clears a path for stronger research questions.
But the practice rarely matches the promise. Researchers recover effect sizes from inconsistent papers, reconcile coding categories, debug scripts, redraw plots, and repair exports. These tasks are necessary, but they are not where scholarly judgment should be consumed.
AI4Meta is Dr. Feng's attempt to move researchers back toward the level of ideas: research questions, hypotheses, conceptual framing, inclusion logic, interpretation, and manuscript decisions.
The platform treats agentic AI as an accountable research apparatus. Agents help with search, screening, extraction, coding, reliability checks, effect-size harmonisation, analysis, and reporting, while the researcher remains responsible for the final scholarly choices.
Make rigorous review work teachable, traceable, and operational.
Method before automation
Agents stay attached to protocol, screening, extraction, analysis, and report objects instead of floating as free-form chat.
Human accountability
Important research choices stay visible, reversible, and attributable to the team that signs the work.
Provenance as evidence
Cells, claims, analyses, and reports are designed to point back to source evidence and review decisions.
Model pluralism
Different model families are routed to the work they are best suited to perform, then checked against the method.
Humans own the ideas. Agents handle the apparatus.
The platform draws a deliberate boundary between scholarly judgment and technical execution. The boundary is visible, adjustable, and logged.
Human-owned conceptual work
Research questions, hypotheses, theoretical framing, inclusion rules, codebook intent, final interpretation, and manuscript claims.
Agent-assisted technical work
Search expansion, deduplication, screening drafts, extraction, reliability checks, effect-size harmonisation, plots, and report scaffolding.
From protocol to report without losing the evidence trail.
Workflow stages are treated as research objects rather than loose prompts. That keeps the work easier to inspect, correct, and reproduce.
- 01Frame the protocol and question
- 02Search, import, and deduplicate studies
- 03Build the codebook and extract evidence
- 04Check reliability, curation, and audit trail
- 05Run synthesis or meta-analysis
- 06Export reports, media, and reproducible records
Built with many models, not one magic model.
AI4Meta has been developed and tested through a broad OpenClaw-style model bench. The intended operating principle is pragmatic: use stronger models when methodological care is required, and use faster models for routine work.
- OpenAI and Codex through OAuth for high-stakes reasoning and development orchestration
- Claude, Gemini, Kimi, GLM/Z.ai, Qwen/DashScope, MiniMax, DeepSeek, NVIDIA NIM, OpenRouter, and local OpenAI-compatible endpoints
- Fast and non-premium models for routing, triage, formatting, and repetitive checks
- Stronger models reserved for architecture, methodological judgment, debugging, and difficult SR/MA reasoning
A living research platform, developed through iterative human-AI collaboration.
- Application
- Next.js, React, Tailwind CSS, FastAPI, SQLAlchemy, Postgres, Redis, PgBouncer, workers, and Docker Compose.
- Agent runtime
- OpenClaw-style orchestration, skills, model routing, job events, project-aware tools, and reviewable outputs.
- Research scope
- Meta-analysis, systematic review, scoping review, content analysis, bibliometrics, extraction, reliability, and reporting.
- Governance
- Admin-governed modules, tier-aware access, audit trails, MCP, API controls, and tenant-provided model options.
Use AI4Meta as a serious research instrument.
Explore the documentation, API, and MCP guides for the public AI4Meta service at ai4meta.net.