NeoHive
Self-hosted context and memory server for coding agents. Ranked retrieval over codebase and team knowledge; runs in your own infrastructure.
เอกสาร
Value per token
4.4 x more
Relevant results per dollar
Accuracy
87 % improvement
More of your own code found
Agentic results
3.6 x better
Tasks passed on all attempts
Figures from two published benchmarks, each with its full method: four models graded blind on proprietary codebases, and 97 τ³-Banking tasks run five times each.
01
Connect your data
Shared persistent context with zero maintenance
Connect it once and it ingests your team's repos, docs and tickets on its own, then keeps up as they change.
- Stays inside your network
- Curates itself
- Useful on day one
Connect agents
$ claude mcp add neohive
Connect through MCP. Every agent on your team is then reading the same memory.
02
Collaboration platform for agents
Every agent collaborating on the same team context
Claude Code, Codex and Cursor connect to the same hive over MCP, with no per-tool setup.
- Fix it once, not four times
- Same answer in every tool
- New hires start informed
Capture decisions and conventions
The reasoning behind a choice is the first thing lost, so it is kept beside the choice.
03
Self-improving memory
The more you run it, the better it answers
Retrieval reranks on what your team actually opens, so the useful answers surface first.
- What gets used, strengthens
- What goes stale, fades
- Prunes itself
Retire what stops being used
A memory that stops being retrieved stops surfacing, so the pruning happens on its own.
convention repo
hi, Neo here your team's sources all agents write here learning as we go see you in the hive found it I'll remember that
One container. Every hive, in context.
A dashboard for every hive, see your connected team data, gain insights on your shared memories and conventions.
Overview
- Rate limiting the auth endpoint docs/auth.md · updated 2d ago
- Error handling convention docs/conventions.md · updated 6d ago
- Billing API, v2 migration docs/billing.md · updated 3w ago
Your code never leaves your infrastructure.
NeoHive runs entirely in your own environment.
Local container
Runs in your VPC or on your machine. Embedding and retrieval both happen locally inside it.
No outbound calls
The container makes no external requests of its own. The only external traffic is the agent API calls your team already makes.
On-premises by default
Deploy on your own infrastructure from day one. Air-gapped and VPC-compatible for regulated environments.
Your existing permissions
Respects the access controls you have. Engineers see answers derived only from content they are allowed to read.
Local container · No outbound data · Embedding stays local · VPC-compatible
The only external traffic is the agent API calls your team is already making, so adding NeoHive opens no new path out of your network.
[
Engineering
Why MD Files Are Not Enough
CLAUDE.md and AGENTS.md are a good place to start and a poor place to stay. What the evidence says, the five ways they break, and what a context layer does differently.
Dean McGregorChief Technology Officer
](https://www.neohive.ai/blog/why-md-files-are-not-enough/)[
Insights
What Is a Context Layer for AI? Definition, History, and Why It Matters
A context layer is the persistent, governed retrieval tier that decides what your AI tools see. Here is what it means, where the idea came from, and how NeoHive fits.
Ralf HuuckFounder
](https://www.neohive.ai/blog/what-is-a-context-layer-for-ai/)[
Engineering
1.6x More Tasks Solved, 4x Cheaper Than Frontier Models: Agents With a Context Layer
With NeoHive the same model solved 1.6x more unseen tasks, at 1.2x lower cost per correct answer than GPT-5.6 Sol and up to 6.8x lower than Claude Opus 5.
Nader AwadLead AI Engineer
](https://www.neohive.ai/blog/tau-context-layer-agent-bench/)
Start with one repository.The memory grows from there.
Free to get started · No credit card required · Enterprise pilots available