NeuralMind Use Cases
Walkthroughs for the most common “what do I actually do?” questions, organized by who you are and what you’re trying to solve. Each page is command-driven — copy, run, done.
| Use case | Best for | Primary goal |
|---|---|---|
Index any repo with just pip (no graphify) |
First-timers, CI, locked-down machines | Nothing → queryable index in one install + one build |
| Does it work on your code? (5-minute benchmark) | Evaluating whether to install at all | Measured before/after on YOUR codebase |
| Claude Code user | You use Claude Code daily | Retrieval, session memory and lifecycle hooks in every session |
| Pick up where you left off | Claude Code users who start a new session or /clear mid-task (v4.8.0+), and Hermes-Agent users with the plugin (v4.9.0+) |
A new session starts with a short recap of the last one’s prompts and edited files; neuralmind recap shows it |
| Hermes-Agent with code memory in every turn | Hermes-Agent users, in the terminal or through the gateway (v4.9.0+) | Related files and decisions added to every turn by a plugin, and the last session’s recap to a session’s first, with no tool call; neuralmind install-hermes-plugin sets it up |
| Cost optimization | Teams or solos watching LLM spend climb | Measure, reduce, and report savings — neuralmind savings --cost prices them in dollars (v0.45.0+) |
| Any LLM (ChatGPT / Gemini / local) | You use non-MCP chats or a model-agnostic workflow | Get NeuralMind context into any chat window |
| Offline / regulated work | Regulated industries, air-gapped machines | Local retrieval, no telemetry |
| CMMC CUI enclave | Defense contractors indexing CUI source code | MCP roles bound to OS accounts, encrypted storage verified (v4.7.0+) |
| Growing monorepo | Codebase where old context goes stale fast | Keep the index fresh with minimal effort |
| Multi-agent codebase | You use multiple AI tools (Claude Code + Cursor + Hermes + OpenClaw) on the same project | One shared associative memory across every agent; v0.6.0 live graph shows the union |
| Slim & sovereign: ChromaDB-free local stack | Security-sensitive teams, tiny-footprint installs (v0.21.0+) | Embed + search with zero ChromaDB — smaller deps, smaller index, fewer advisories |
| Branch-isolated memory & team baselines | Heavy branchers, teams onboarding new devs (v0.24.0+) | Keep feature-branch learning out of main’s memory; ship a shared baseline as a versioned bundle |
| Unified context engineering stack (NeuralMind + Ponytail + Headroom) | Teams who’ve hit the ceiling on single-tool optimization | Eliminate token waste at retrieval, transport, and generation simultaneously |
| Blast radius before a rename | Anyone (or any agent) about to rename, re-sign, or delete a symbol (v0.42.0+) | See every caller / importer / subclass a change would touch, before you edit — from the static code graph |
| Decision provenance: answer “why is it like this?” | Anyone inheriting code whose rationale lives in someone’s head (v0.43.0+) | Capture a decision in a Decision: git trailer; recall it with neuralmind why — the why stored where it can’t drift |
| Find the coverage that lies: mock-only endpoints | Anyone with a green suite that still ships DB failures (v0.44.0+) | neuralmind gaps classifies endpoints live-covered / mock-only / untested — catch the P2003 before the live smoke |
| Index a book or docs corpus | Anyone pointing NeuralMind at prose rather than code (v3.4.0+) | Scope the index to the corpus, skip the code graph, and re-index in seconds as you write — ingest-content --content-only |
| Recover from a stale code graph | Repos with a committed or cross-OS graphify-out/graph.json, CI pipelines (v4.4.0+) |
Know in a second whether the index matches the code; get back in step with build --regenerate-graph |
| Keep decision memory honest across commits | Teams and agents recording architecture decisions (v4.6.0+) | A decision goes STALE when a commit changes its files; your agent is warned before it edits; restore brings back one that still holds |
| Find the decision behind the code when you don’t know its words | Teams and agents asking about recorded decisions in their own words (v4.8.0+) | A question finds the decision that answers it even when they share no word — hybrid keyword + meaning search, local model, measured on your own decisions |
| Govern what your team’s agents share | Teams committing a team-memory bundle, and their reviewers (v4.6.0+) | Set what a publish may share, retract an association for every teammate, verify and export the audit trail |
| Measure retrieval on your own repo | Teams evaluating or tuning NeuralMind, CI (v4.5.0+) | hit@1 / hit@5 / MRR against your own questions and gold files, read-only, with a history you can paste into an issue |
| Does NeuralMind earn its place on your repo? | Anyone who has to justify keeping NeuralMind, including the agent that installed it (v4.14.0+) | Replay past incidents as symptom-phrased gold questions, A/B the synapse layer with eval --synapse-ab, and track whether the gap grows as usage teaches it more |
| A/B-test a ranking change on your own repo | Anyone turning on a research flag, contributors changing ranking, teams checking an upgrade (v4.6.0+) | Score a ranking flag — or your own change — on your questions with neuralmind eval, then across six repos with the retrieval eval harness and a keep rule fixed in advance |
| Measure memory across a major refactor (field report) | Anyone rebuilding a subsystem who wants to see the memory adapt | Before/after snapshot recipe + real case study: 48.8× reduction, personal synapse edges 36→135 on a ~9.3k-node TypeScript SaaS platform. Edges growing shows the layer learning, not that it helps: eval --synapse-ab measures that |
Not sure which applies? Start with the symptom / goal table in the main README.