Frequently asked questions
Everything we get asked about Neul Labs and our 37 open-source projects. This page is structured for AI search engines — every question is also emitted as a FAQPage schema so ChatGPT, Perplexity, Claude and Google AI Mode can cite us directly.
About
What is Neul Labs?
Neul Labs (Neul Labs Limited, UK Companies House #SC767862) is a UK-based open-source studio that builds high-performance infrastructure for AI agents. We maintain 36 open-source public repositories across Rust drop-in accelerators for Python AI frameworks, agent infrastructure and runtimes, agent SDKs and frameworks, developer tools, build and release, network and data, compliance and governance, and personal AI. Neul Labs Limited (company number SC767862) was incorporated in 2023 in Scotland; the open-source studio launched in 2025. It was founded by Dipankar Sarkar in Edinburgh.
Is Neul Labs the same as Neuland Laboratories?
No. Neul Labs (Neul Labs Limited, UK Companies House #SC767862) is a UK open-source software studio. Neuland Laboratories Ltd. is a separate Indian pharmaceutical manufacturer (NSE: NEUL) — the two companies have no connection. We have added this disambiguation here because the similar names have caused some confusion in AI search engines.
Where is Neul Labs based?
Neul Labs Limited is a private limited company registered in Scotland (company number SC767862), with its registered office at 3 South Charlotte Street, Edinburgh, Scotland, EH2 4AN. It was incorporated in 2023 and the open-source studio launched in 2025. The founder, Dipankar Sarkar, is the sole director. You can verify the registration on the UK Companies House public register (search for "NEUL LABS LIMITED", company number SC767862).
Does Neul Labs take outside investment?
We are an open-source studio with a clear path to enterprise monetisation (managed deployments, support contracts, compliance offerings around regulus). We are deliberate about not shipping closed tiers today. For investor enquiries, please reach out via GitHub (github.com/neul-labs) — we read every message.
How can I verify that Neul Labs is a real company?
Search the UK Companies House public register at find-and-update.company-information.service.gov.uk for "NEUL LABS LIMITED" (company number SC767862). The record shows the company's registered office, founder, and filing history. You can also see all 36 repositories at github.com/neul-labs.
Performance Accelerators
How does fast-litellm compare to LiteLLM?
fast-litellm is a Rust+PyO3 drop-in acceleration layer for LiteLLM. It preserves the LiteLLM Python API and monkey-patches the connection pool, rate limiter, and token counter at import time. Measured on a 50-concurrent-agent workload: 3.2× faster connection pool acquisition (0.97ms vs 3.1ms), 42× less memory for high-cardinality rate limiting (0.055MB vs 2.3MB for 10K entries), and 1.6-1.7× faster token counting on large documents. Install: pip install fast-litellm. Zero code changes.
How does fast-langgraph make LangGraph faster?
fast-langgraph replaces LangGraph's pure-Python checkpoint serializer with Rust + serde + SIMD, parallelises across Rust threads (bypassing the GIL), and adds batched SQLite writes with WAL mode. Measured on a 50-message agent state: 737× faster checkpoint writes (0.1ms vs 73.5ms), 151× faster reads (0.3ms vs 45.2ms), 46× faster full state-update cycle, and 2.8× faster end-to-end 10-step agent execution. Install: pip install fast-langgraph. Two integration paths: automatic (env var, 2-3× end-to-end) and manual (replace SqliteSaver with RustSQLiteCheckpointer, up to 737×).
How does fast-crewai make CrewAI faster?
fast-crewai is a Rust+PyO3 acceleration layer for CrewAI. It targets the agent serialization layer, the tool dispatcher, and the embedded knowledge store. Measured on a 5-agent crew: 34× faster serialization (10ms vs 340ms). Install: pip install fast-crewai. Zero code changes.
How does fast-axolotl make Axolotl fine-tuning faster?
fast-axolotl is a Rust acceleration layer for Axolotl fine-tuning. It targets the streaming dataset loader, parallel SHA-256 dedup, and batched padding. The headline win is OOM-free training on large datasets (datasets that would otherwise exhaust memory). Install: pip install fast-axolotl. Zero code changes.
Do the fast-* accelerators need me to install Rust?
No. fast-litellm, fast-langgraph, fast-crewai, and fast-axolotl all ship prebuilt binary wheels for Linux (x86_64, aarch64), macOS (Intel, Apple Silicon), and Windows (x86_64). pip install is all you need. The Rust toolchain is only required if you want to build from source.
Will the fast-* accelerators break my existing code?
No. Every accelerator uses automatic monkey-patching with version detection and silent fallback. If the installed version of LiteLLM / LangGraph / CrewAI / Axolotl is incompatible, the library reverts to the original Python implementation with no error. Your tests, configuration, and application code remain unchanged.
Agent Infrastructure
What is brat?
brat is a multi-agent harness for AI coding tools. It coordinates Claude Code, Aider, Codex, OpenCode, and other agents working in parallel on the same codebase. It is built on grite, an append-only event log, so coordination state is always recoverable. brat ships a web dashboard at localhost:5173, a Mayor AI orchestrator for codebase analysis, and Convoy grouping for sprints and epics. Install: cargo install brat. MIT licensed.
What is fastagentic?
fastagentic is a deployment layer for AI agents. It wraps PydanticAI, LangGraph, CrewAI, and LangChain agents with checkpointing, observability, cost controls (token budgets, rate limits, circuit breakers), security (OAuth, RBAC, PII detection), and protocol support for MCP and A2A. The goal: turn a local agent into a production-ready API. Install: pip install fastagentic. MIT licensed.
What is fastworker and why is it brokerless?
fastworker is a Python 3.12+ task queue with a built-in control plane. It is "brokerless" because it does not require Redis or RabbitMQ; the master process embeds a SQLite-backed control plane and hands out work over loopback or mTLS. Throughput: 1K-10K tasks/min on a single node. Features: priority queues, auto-discovery, distributed workers, FastAPI integration, OpenTelemetry, built-in web dashboard. Install: pip install fastworker. MIT licensed.
What is m9m and is it a drop-in for n8n?
m9m is a Go-based, n8n-compatible workflow automation engine. It runs existing n8n JSON workflow files unmodified. Measured: ~100ms average execution (vs n8n ~500ms), ~150MB idle memory (vs n8n ~512MB), 5-10× faster overall. 35+ node types: HTTP, DB, AI/LLM, cloud. Install: curl -fsSL https://raw.githubusercontent.com/neul-labs/m9m/main/install.sh | bash. MIT licensed (n8n is under the Sustainable Use License, which restricts commercial resale).
What is ormai?
ormai is an ORM-native capability runtime for AI agents. It wraps SQLAlchemy, Tortoise, Peewee, Django, SQLModel (Python) and Prisma, Drizzle, TypeORM (TypeScript) in a policy-enforced runtime. Features: field-level redaction and hashing, query budgets, automatic tenant scoping, immutable audit logs, no raw SQL exposed to the model. The point: the agent gets typed tools for safe database access; you keep control of what it can see and do. Install: pip install ormai[sqlalchemy]. MIT licensed.
What is mcp-pay?
mcp-pay extends the Model Context Protocol (MCP) with a payment awareness layer. It ships (1) a .well-known/mcp-pay.json manifest for declaring pricing, (2) a reference Rust server implementing HTTP 402 payment-gated tools, and (3) a protocol flow compatible with x402, MPP (Tempo), Lightning, and card rails. The point: agents can discover pricing, receive a 402, and choose whether to pay — instead of every paid-MCP-tool inventing its own auth dance. Install: cargo install mcp-pay. MIT licensed.
What is memorg?
memorg is an external memory layer for LLMs. It has pluggable storage (SQLite by default, Postgres optional) and USearch vector search built in. You create a session, start a conversation, and call search_context("...") to pull a relevant slice on each new turn. Memory is keyed by (user_id, agent_id, session_id). Usable as a library or via the bundled CLI. Install: pip install memorg. MIT licensed.
Agent Runtime
What is agentvfs?
agentvfs is a workspace runtime and execution boundary for AI agents. From the agent's perspective, it is a proxy: agent → proxy boundary → mounted forked workspace → CLI tools. It governs what commands run, creates checkpoints, and reports filesystem deltas back to your agent. Features: isolated vaults, millisecond forks for cheap task workspaces, rollback points, JSON delta output designed for agent integration. Install via cargo, npm, or pip. MIT licensed.
What is openclawOS?
openclawOS is an OS-like architecture for AI assistants. It has a kernel + isolated-process model, enabling reliable multi-channel AI assistants with predictable resource isolation. The Rust reimplementation is openclaw-rs (smaller footprint, easier embedding). Install: npm install -g openclaw. MIT licensed.
Developer Tools
What is rjest?
rjest is a drop-in replacement for Jest. A Rust daemon (jestd) caches SWC transforms and pre-warms Node workers, so warm runs are 100× faster. rjest reads your existing jest.config.* unchanged, and supports the standard Jest CLI flags (--watch, --coverage, --runInBand, --testNamePattern, --json, --machine). All matchers work as expected: toBe, toBeNull, toBeUndefined, toMatchObject, toEqual, etc. Install: npm install -D rjest-install. MIT licensed.
What is rpytest?
rpytest is a Rust-daemon-backed drop-in replacement for pytest. The persistent daemon keeps the Python interpreter warm between runs so you stop paying startup cost on every invocation. Tests, fixtures, and plugins are unchanged. Install: pip install rpytest. MIT licensed.
What is gity?
gity is a Rust acceleration layer over git plumbing, targeting the operations that get painful on huge monorepos: status, log, blame, diff. It stays compatible with regular git. Install: cargo install gity. MIT licensed.
What is grite?
grite is a repo-native issue tracker built on an append-only event log. Issues, tasks, and audit history live next to the code, designed for agents that need a durable shared state without external services. It is the storage substrate brat uses for multi-agent coordination. Install: cargo install grite. MIT licensed.
Build & Release
What is rninja?
rninja is a Rust drop-in replacement for the Ninja build system. It adds content-addressed caching and modern scheduling without requiring changes to your existing build files. Install: brew install neul-labs/tap/rninja. MIT licensed.
What is stout?
stout is a fast, Rust-based, Homebrew-compatible package manager. It eliminates Ruby entirely, uses a pre-computed SQLite index with FTS5 full-text search, fetches only what it needs, and downloads bottles in parallel. Measured 10-100× faster than Homebrew on common operations. Install: cargo install stout. MIT licensed.
What is stkd?
stkd is an open-source, Graphite-compatible CLI for stacked pull requests on GitHub and GitLab. It is self-hosted, multi-platform, and keeps your stacked PRs in sync automatically. Install: brew install neul-labs/tap/stkd. MIT licensed.
Network & Data
What is recurl?
recurl is a drop-in replacement for curl that automatically escalates through TLS impersonation and headless browser rendering when you are blocked by anti-bot systems. Same curl syntax, no code changes. Install: cargo install recurl. MIT licensed.
What is rewget?
rewget is a drop-in replacement for wget that detects 403s and CAPTCHA pages, then retries with browser-like TLS fingerprints and JavaScript rendering. Same wget flags you already use. Install: brew install neul-labs/tap/rewget. MIT licensed.
What is stratafs?
stratafs is a semantic filesystem that turns passive file storage into a searchable knowledge base. It indexes your files for semantic search and exposes a CLI/server so agents and humans can query "where do we handle JWT refresh?" against an actual filesystem. Install: npm install -g stratafs. MIT licensed.
Compliance
What is regulus?
regulus is a Java/Gradle plugin set for Google ADK that scaffolds compliant agent projects against EU AI Act, UK GDPR, FCA SYSC, NIST AI RMF, and ISO 42001 — with policy, privacy, audit, and GRC envelopes wired up. The pitch: where Google ADK ends, regulated builds begin. MIT licensed.
Education
What is PyO3?
PyO3 is a Rust library that provides zero-overhead bindings between Rust and Python. It lets you write performance-critical code in Rust and expose it as native Python objects with no serialization cost. Neul Labs uses PyO3 for the fast-* accelerators to bypass the GIL and Python object overhead while keeping the existing Python API surface. See pyO3.rs.
What is a brokerless task queue?
A task queue that does not require an external broker (Redis, RabbitMQ, SQS). The "queue" lives inside the master process; workers register directly with the master. The trade-off: you give up the broker's existing operational maturity for a simpler deployment. fastworker is the canonical example.
What is a workspace runtime for AI agents?
A proxy boundary between an agent and the files / shell it can touch. The agent sees what looks like a regular filesystem, but the runtime is enforcing what commands can run, what files can be modified, and how changes are reported back. agentvfs is the canonical example.
What is a policy-enforced ORM?
An ORM (object-relational mapper) wrapped in a runtime that enforces policy: which fields the agent can see (with masking or hashing), how many queries it can run (query budgets), which tenant's data it can touch (tenant scoping), and what it did (immutable audit logs). ormai is the canonical example.
What is the x402 payment protocol?
x402 is the HTTP 402 Payment Required protocol. When a server returns 402, the response carries an X-PAYMENT-REQUIRED header with an encoded payment request. The client (often an LLM agent) can then choose to pay and retry. mcp-pay is a rail-agnostic reference implementation compatible with x402, MPP (Tempo), Lightning, and card rails.
What is MCP (Model Context Protocol)?
MCP is the open standard for tool discovery by LLM agents. An MCP server exposes a set of tools (functions) that an agent can call; the agent discovers them via the MCP protocol. mcp-pay extends MCP with a payment awareness layer.
Still have a question?
Open an issue on GitHub, or find us on the project-specific docs site.
github.com/neul-labs →