Collected molecules will appear here. Add from search or explore.
Gerolamo indexes GitHub repos, arXiv papers, and HuggingFace models — then scores each for defensibility, threat profile, and composability. Search the corpus, discover sleepers, compose intelligence with AI, design meta molecules that evolve through generations, compare foundation models with Mentis, personalize your feed, and connect your agents for autonomous research workflows via MCP.
Find any technology across three platforms in one query.
Semantic search understands what you mean, not just what you type. Search for 'autonomous drone swarm' and find repos, papers, and models — even if they use different terminology. Keyword mode for exact matching. Filter by corpus, date, and score.
A robotics engineer searches 'SLAM navigation indoor' and finds a GitHub repo, an arXiv paper proposing a novel approach, and a HuggingFace model trained for it — all scored and ranked.
Know what survives and what gets replaced.
Every molecule is scored 1-10 for defensibility by AI analysis. Threat profiles show Platform Domination Risk (could AWS absorb this?), Market Consolidation Risk (will the market shrink?), and Displacement Horizon (how long until a competitor replaces it?).
A PE analyst evaluating an acquisition target checks its defensibility score (7/10) and sees PDR:HIGH — meaning a big platform could absorb it within 1-2 years. That changes the valuation conversation.
Discover what nobody else is watching yet.
Sleeper mode filters to high-defensibility, low-traction molecules — the hidden gems. 'Surprise Me' shows a random sleeper with one click. These are the repos and papers that score well but haven't been discovered by the crowd.
An investor clicks 'Surprise Me' every morning and discovers a 3-day-old repo by a former DeepMind researcher with 23 stars and a defensibility score of 8. Early signal before anyone else.
Browse the full corpus with powerful filters.
Sort by score, newest, velocity, or traction. Filter by corpus, novelty type, composability, and minimum score. Signal finder mode surfaces high-score molecules with low traction. Score distribution histogram shows the shape of each topic.
A CTO browses all 'breakthrough' novelty molecules in the GitHub corpus, sorted by velocity. Finds 3 repos gaining stars faster than anything else in their domain.
AI-generated intelligence summaries across technology domains.
Overview of 66+ tracked technology domains with entity counts, defensibility averages, and AI-written narratives. Trends tab shows how domains evolve over time. Capabilities tab maps which technical capabilities are mature, emerging, or nascent.
A strategy team checks the Conspectus for 'quantum computing' and sees: 195 entities, avg defensibility 1.9, 3 new this cycle. The narrative explains that most quantum repos are educational — real infrastructure plays are rare and defensible.
One number that tells you if something's happening.
A daily composite score measuring the pulse of technology emergence. Computed from new molecules, breakout events, sleeper density, ecosystem velocity, and cross-platform activity. When the GIX spikes, multiple technology domains are moving simultaneously.
The GIX jumped from 42 to 78 overnight. An analyst checks and finds: 3 breakouts in the AI agent space, 12 new sleepers in robotics, and a surge of cross-platform activity in post-quantum cryptography. Something is happening.
Compare foundation models on price, capability, and market risk — not just benchmarks.
Mentis tracks pricing, capabilities, context windows, and benchmark scores across foundation models from OpenAI, Anthropic, Google, Meta, and others. The Economics tab plots cost vs. capability on a scatter chart. The Timeline tab shows how pricing evolves over time. The Capability Matrix maps which models support which features. Domination Risk analysis shows where one provider could lock in a market segment.
A platform team choosing between GPT-4o and Claude Sonnet opens Mentis, compares them side-by-side on pricing, context window, and vision support, then checks Domination Risk to see if either provider is likely to corner the market for their use case.
Collect molecules and compose them into actionable outputs.
Add up to 20 molecules to your workspace, then use AI to compose them. Build mode generates a SPEC.md with architecture and commands. Research mode writes an investment thesis. Augment mode creates an integration brief with a task backlog. Compare mode produces decision-ready trade-off analysis.
An engineer collects 5 MCP server implementations, clicks 'Build', and gets a complete SPEC.md with project structure, dependencies, and build commands. They paste it into Claude Code and start building.
Save, rename, share, and download your AI-generated outputs.
Every workspace composition is automatically saved. Browse your history, rename them, download as .md files (Build mode downloads as SPEC.md), or share via public link. Your compositions become reusable intelligence assets.
A team lead composes a Research brief on the LLM inference space, shares the public link with their investment committee, and downloads the .md for their internal wiki.
Design synthetic entities that compose with real molecules and evolve through generations.
Meta molecules are ideas you author — synthetic entities that live alongside real repos, papers, and models in the corpus. Create one by selecting parent molecules for inspiration, describing what you want to build, and letting Gerolamo score its defensibility. Meta molecules track lineage: every derivative knows its ancestors, and you can trace the full genealogy of an idea. When you build the real thing, connect it to realize the meta molecule and close the loop.
A researcher searches for GPS-denied navigation approaches, finds 4 relevant repos, and creates a meta molecule called 'HYDRA-NAV' that fuses their ideas into a novel degradation-resilient architecture. Gerolamo scores it 7/10. Three months later, she builds the actual repo and realizes the meta molecule — now it's linked to the real entity with full lineage tracking.
Track molecules over time and see what changed.
Bookmark any molecule for quick access. The 'What Changed' tab shows trajectory diffs — traction deltas, velocity changes, and status labels (breakout, growing, stable, declining). Portfolio-style monitoring for your thesis.
An analyst bookmarks 8 molecules in the autonomous agents space. A week later, the 'What Changed' tab shows one went from 50 to 400 stars (breakout) and two are declining. Time to update the thesis.
Let your AI agents use Gerolamo as their intelligence layer.
23 MCP tools for search, discovery, composition, and counter-intelligence. score_stack analyzes weakest-link defensibility across your tech stack, explain_score unpacks the full reasoning behind any score, find_alternatives surfaces replacement candidates by embedding similarity, and suggest_tools recommends which Gerolamo tools to call for any task. REST API available for any HTTP client. The MCP server is open source at github.com/adjective-rob/gerolamo-mcp.
A Claude Code agent connects to Gerolamo via MCP, searches for 'protein folding', finds 3 sleeper repos, composes a Build spec fusing them together, and begins implementation — all autonomously.
Let AI agents conduct full research workflows — not just answer questions.
Gerolamo's MCP server isn't just a search endpoint. Agents can chain tools together to conduct multi-step research: search for a domain, discover sleepers, analyze competitive landscapes, score tech stacks for weakest-link risk, create meta molecules from their findings, compose build specs, and save the results — all in a single autonomous session. The agent reasons about what tools to use based on the task, not a script.
You tell Claude Code: 'Research the autonomous drone navigation space and design something novel.' The agent calls search_intelligence, finds 12 relevant molecules, runs analyze_competitive_landscape to map the field, identifies a gap using explore_connections, creates a meta molecule via submit_meta_molecule, composes a Build spec with compose_molecules, and saves it with save_composition. You review the output.
Pre-built workflows that show agents how to chain Gerolamo tools together.
The gerolamo-workflows repo contains documented, copy-paste-ready agent workflows for common research tasks. Each workflow specifies which MCP tools to call, in what order, and what to do with the results. Workflows cover landscape analysis, competitive scouting, build planning, tech stack audits, and more. Open source at github.com/adjective-rob/gerolamo-workflows.
An engineer wants to audit their dependency stack for defensibility risk. They grab the 'stack-audit' workflow, paste it into Claude Code, and the agent runs score_stack, finds two weak links, calls find_alternatives for each, and produces a migration plan — no prompt engineering required.
Run the MCP server yourself or contribute new tools.
The Gerolamo MCP server is fully open source. Clone it, run it locally, add custom tools, or connect it to your own infrastructure. The repo includes setup instructions, tool documentation, and examples for integrating with Claude Code, Cursor, and other MCP-compatible clients.
A team forks gerolamo-mcp, adds a custom tool that cross-references Gerolamo intelligence with their internal JIRA board, and deploys it as their engineering team's private research agent.
Get notified when new molecules match your criteria.
Set up semantic queries that run on a schedule. When new molecules appear that match your criteria and score threshold, you get notified. Never miss an emerging tool in your domain.
A security researcher sets an alert for 'zero-day exploit detection AI' with a minimum score of 5. Two weeks later, a new repo appears that matches — they're the first to know.
Tell Gerolamo what you care about and get results tuned to you.
Set your interests — corpora, source types, technology domains, and maturity preferences — in Settings. The 'For Me' toggle on search reranks results based on your profile, surfacing what matters to you first. Global mode still gives you the full unfiltered corpus.
A robotics engineer sets their interests to GitHub repos in 'autonomous agents' and 'robotics' with breakthrough novelty. Now every search with 'For Me' enabled prioritizes those domains and filters out noise from unrelated fields.
See the full intelligence neighborhood around any molecule.
Every molecule exists in context. Explore Connections shows 5 rings: the entity itself, the creator's other work, semantically adjacent projects, shared capability space, and topic-level trends. With AI synthesis enabled, Gerolamo identifies gaps in the landscape and suggests specific capabilities you could build by composing available molecules.
An engineer explores a SLAM library and discovers the creator also built 4 other robotics tools, there are 3 semantically similar projects they didn't know about, and the AI suggests combining two of them into a new indoor navigation capability that nobody has built yet.
Understand what could displace any technology — and how to defend against it.
Threat mode on Explore Connections maps competitive risks for any entity. Identifies which frontier labs, well-funded startups, or platform plays could make a technology obsolete. Estimates displacement timelines. Provides a defense playbook with specific strategic moves to build lasting defensibility.
A founder building an AI code review tool runs threat analysis on their space. Gerolamo identifies that GitHub Copilot is 6 months from shipping native code review, and recommends pivoting to a compliance-focused niche that platforms won't enter.