# Lab Notes: GPT-Researcher - Multi-Agent Researcher with Citations > Explore Infainite's Labs team evaluation of GPT-Researcher as a multi-agent system for research and content generation, including deployment challenges and future possibilities. **Canonical:** https://infainite.ai/blog/gpt-researcher **HTML:** https://infainite.ai/blog/gpt-researcher **Source:** src/data/blog/gpt-researcher.md --- > **Labs experiment** > > Can an open-source multi-agent researcher produce cited, long-form output reliably enough to demo to customers? > > *We deployed GPT-Researcher to find out.* Infainite Labs ran a structured evaluation of [GPT-Researcher](https://github.com/assafelovic/gpt-researcher): deploy, stress-test orchestration, and document what is production-adjacent vs demo-only. ## At a glance | | | |--|--| | **Project** | Open-source multi-agent research and report generation | | **Stack** | LangChain graph orchestration, containerised deployment | | **Standout strength** | Long-form output with citations | | **Main friction** | Azure fit, browser quirks, expert-first UI | | **Lab verdict** | Strong research accelerator with human editorial gate, not unattended publishing | ## In this article 1. What we tested and how. 2. Findings by theme (tech, UX, repo hygiene). 3. Pros and cons scorecard. 4. What we would do differently on the next iteration. ## Experiment design | Phase | Activity | |-------|----------| | **Discover** | Map agent roles, dependencies, stable branch | | **Deploy** | Containerised instance; patch browser inconsistencies | | **Evaluate** | Research tasks on technical topics; measure citation quality and latency | | **Demo prep** | Simplified UI narrative for non-expert audiences | ## What worked ### LangChain graph orchestration A lead agent coordinates retrieval, analysis, and citation agents. Long reports stay **coherent** because steps are explicit, not one bloated prompt. ### Cited long-form output For marketing, policy, and technical briefs, the **research + cite** loop is the headline capability, editors still required, but first drafts arrive faster. ### Containers as equaliser Docker packaging made dependencies reproducible across laptops and demo environments, critical when Azure-native deploy hit mismatches early. > **Key takeaway:** Multi-agent value showed up in **structure**, not just model size. ## What hurt | Issue | Impact | Mitigation we used | |-------|--------|-------------------| | **Azure friction** | Slower first deploy | Containers + revised networking | | **Agent latency** | Long end-to-end runs | Parallelise retrieval where possible | | **Expert UI** | Steep for business viewers | Guided demo script + simplified surface | | **Browser variance** | Broken flows in some browsers | Targeted patches for demo path | | **Branch clarity** | Hard to pick stable baseline | Pin commit + internal fork notes | ## Pros and cons scorecard | Pros | Cons | |------|------| | Strong citation discipline | Orchestration ops overhead | | Modular agent roles | Needs ML-adjacent operator | | Active open-source community | UI not ready for general staff | | Good fit for Labs storytelling | Not turnkey SaaS for clients | ## Future directions we are watching | Direction | Why it matters | |-----------|----------------| | **Research APIs** | Productise agent teams behind one endpoint | | **Editorial workflows** | Human-in-the-loop approval before publish | | **“AI browser” pattern** | Fetch + summarise + cite while browsing | ## Next steps for Infainite Labs - Publish build notes and deployment checklist. - Run live demo with real-time citations. - Compare against other open-source agent frameworks on the same benchmark tasks. ## References - [GitHub: GPT-Researcher](https://github.com/assafelovic/gpt-researcher) - [LangChain documentation](https://python.langchain.com/docs/get_started/introduction) Interested in multi-agent research pipelines for your organisation? [Talk to the team](/contact) about what a governed pilot could look like.