Which AI coding agents does Nexus Prime support?

12+ including Claude Code, Claude Desktop, Cursor, Codex, Windsurf, Cline, Antigravity, Aider, Continue, OpenClaw, and Gemini. Setup is one command per client: nexus-prime setup .

the agent runtime stack · ship with AI, not vibes

Coding agents that
finally ship.

Q: What is Nexus Prime, in one sentence?

Nexus Prime is the agent runtime stack — five blocks (Memory, Routing, Sandbox, Bridge, Audit) that snap onto Claude Code, Cursor, Codex, and Windsurf. Agents remember, stay in bounds, and cost less. One install for your whole stack, no data egress.

One runtime for Claude Code, Cursor, Codex, and Windsurf: memory, routing, safe writes, and audit proof without another model bill.

monthly agent bill $4,300 $1,344 · same code shipped

Install free → Talk to sales

68%

Lower bill

Re-onboard

12+

Agents wired

15.2K

Downloads

› runtime liverouting

8.5msrecall

3.2×cheaper

12+agents

Wired into real agent stacks

IIM Ahmedabad

rezo.ai

IIT Delhi

Qure.ai

orbo.ai

IIT Gandhinagar

IIM Kozhikode

IIM Ahmedabad

rezo.ai

IIT Delhi

Qure.ai

orbo.ai

IIT Gandhinagar

IIM Kozhikode

› Outcomes · [01/09]

What changes in your week.

Three things your team will notice by Friday. Not benchmarks. Outcomes.

−68%

Spend, same work.

Agents stop re-reading your repo every turn. The bill follows.

To resume a task.

Yesterday’s session is one keystroke away. New agent? Same memory.

<24h

Compliance review.

Nothing leaves the box. Every decision is logged and replayable.

› For who · [02/09]

One runtime. Two buyers.

The same five blocks ship for the IC tightening a Cursor session and the platform team rolling agents out across the company. Pick your lane.

{ } For developers & ICs

Your daily agent remembers. Your bill drops.

Install once. Claude Code, Cursor, Codex, and Windsurf inherit memory, routing, and safe edits. No config marathon.

sub-10ms

Recall

3.2×

Cheaper

12+

Agents wired

npm-installable · Mac, Linux, Windows · no API keys

Install free →

☆ For team leads, orgs & enterprise

Run the fleet inside your perimeter.

One policy for Cursor, Codex, Claude Code, and Windsurf. Bring your model gateway. Approve changes before they become real.

Policy consolecentral rules

Replayable auditlogged

SSO + RBACon the roadmap

On-prem & VPCnever leaves the box

Talk to sales → See the contract

› Use cases · [03/09]

Built for the work you ship on Friday.

Four stories teams tell us within a week of installing Nexus Prime.

The founder · burn cut

“My agent was billing me $4k a month to re-read the same files.”

Claude Code → ranked context → model sees 32% of input

Nexus reads what helps and skips the rest. You ship the same code for about a third of the spend.

The team lead · instant resume

“Every Monday my engineers re-onboard the AI to the codebase.”

last week’s session → persistent memory → today, picks up exactly where it stopped

Memory survives the weekend, the IDE, and the agent switch. The first Monday prompt is “keep going.”

The senior · safe refactors

“I can’t let an agent touch main. Last time it broke prod.”

agent edit → staged dry-run → review → PR

Every change is staged and reviewed before it’s real. Main is never touched, rollback is a delete.

The platform owner · one fleet

“Engineering uses five different agents. None of them remember each other.”

Cursor ↔ Codex ↔ Claude Code ↔ Windsurf

One install, one memory, one audit trail. The fleet finally acts like one team.

› Validation · [04/09]

What builders say,
in their own words.

StrictSeal Technical Audit

Nexus-prime has been reviewed and approved.

"We installed the global npm package and verified the CLI toolset ourselves. The MCP server handles memory and token orchestration correctly. GitHub activity shows 27 consistent releases."

u/ultrathink-artr/LLMDevs

"The bounded execution piece is the most valuable part of this architecture - unbounded agents accumulate stale context that silently distorts later decisions...."

rezo.ai EngineeringDirect Message

"We wired Nexus into our agent pipeline and the token savings were immediately measurable. The sandboxed execution model is exactly what we needed for safe multi-agent work."

u/Otherwise_Waver/Adulting

"This is a really interesting direction, control planes for coding agents feel like the missing middle layer between 'chat in an IDE' and actual reliable automation."

u/Time-Dot-1808r/LocalLLaMA

"The core problem is that 'session' as a unit of work does not map to 'task'. A feature spans days, not hours..."

u/ultrathink-artr/LLMDevs

"The bounded execution piece is the most valuable part of this architecture - unbounded agents accumulate stale context that silently distorts later decisions...."

rezo.ai EngineeringDirect Message

"We wired Nexus into our agent pipeline and the token savings were immediately measurable. The sandboxed execution model is exactly what we needed for safe multi-agent work."

u/Otherwise_Waver/Adulting

"This is a really interesting direction, control planes for coding agents feel like the missing middle layer between 'chat in an IDE' and actual reliable automation."

u/Time-Dot-1808r/LocalLLaMA

"The core problem is that 'session' as a unit of work does not map to 'task'. A feature spans days, not hours..."

› The Stack · [05/09]

Five blocks.
One runtime.

Nexus Prime is the agent runtime stack. Each block snaps onto the next. You install one package and your whole agent fleet inherits all five.

Block 01

Memory

Resume yesterday’s task in one keystroke.

sub-10ms recall

Block 02

Routing

Only useful context reaches the model.

3.2× lower spend

Block 03

Sandbox

Every edit stages before it becomes real.

safe by default

Block 04

Bridge

Hand off Cursor → Codex → Claude mid-task.

12+ agents wired

Block 05

Audit

Every decision and patch is replayable.

replayable run log

› The Receipt · [06/09]

Same agent. A third of the bill.

A 20-file plan, run twice. Once raw, once routed through Nexus. The bars are real tokens from our measured 20-file benchmark.

Tokens consumed on a single 20-file planning task

−13,085

tokens / task · 3.2× less

Agent alone: reads everything, every turn.

Agent + Nexus: ranks files, ships only what helps.

Under the hood · full numbers + sources

Measured locally, single process, commodity hardware. External numbers sourced from each project’s published paper or research page. npm test reproduces ours.

Capability	Nexus Prime	Mem0	Zep	SuperMemory
Memory recall p50	8.5ms	~148ms	~513ms	<300ms
Memory store p50	4.6ms	~100ms	~65ms	n/p
Retrieval quality (R@5)	75%	66.9% LoCoMo	71.2% LongMemEval	85.2% LongMemEval
Session cost (10 recalls)	~12.5K fixed	~18K+	~18K+	n/p
Footprint	Local store	Cloud API	Cloud API	Cloud API
Network required	No	Yes	Yes	Yes
Sandbox + audit	Yes	No	No	No

Nexus measured 2026-04-13 on v6.0. External numbers: Mem0 (arXiv:2504.19413), Zep/Graphiti (arXiv:2501.13956), supermemory.ai/research. n/p = not published.

› The Fleet · [07/09]

Every agent in your stack.
One language.

Drop Nexus Prime in once. Every coding agent your team uses starts speaking the same memory, the same routing, the same audit. No glue code. No bespoke integration.

Wired today

Claude Code Claude Desktop Cursor Codex Windsurf Cline Aider Continue Antigravity OpenClaw Gemini JetBrains AI

One command per client: nexus-prime setup <agent>

› Enterprise · [08/09]

Your runtime —
your perimeter.

When the runtime stays inside your network, your compliance review gets shorter. No vector vendor to vet. No model provider to negotiate. No third-party retention policy to chase.

For platform & security teams

Zero data egress. One audit log. Compliance in weeks, not quarters.

Nexus Prime runs inside your developers’ perimeter. Nothing leaves the box. Pair it with your own model gateway and the entire agent fleet operates on infrastructure you already own and have already cleared.

Stays on the box. Memory, routing, audit — all inside the developer’s machine or your VPC.

Pre-flight review. Every change runs as a dry-run before it’s real. Reject and it disappears.

Isolated workspaces. Main branch and shared state are never touched by an agent.

Replayable audit. Every decision, patch, and tool call logged. Compliance gets the receipts.

Twelve agents, one policy. The whole fleet inherits your rules from one install.

Bring your model gateway. Use what you’ve already approved. No third-party retention to chase.

Talk to sales → Full comparison

› Install · [09/09]

Three commands.
You're routed.

Runs on macOS, Linux, and Windows. Node 18+. No API keys.

Step 01 // Install

Pull the package globally from npm.

$npm i -g nexus-prime

Step 02 // Wire your agent

One command per client. Repeat for cursor, codex, windsurf.

$nexus-prime setup claude

Step 03 // Launch

Runtime up. Dashboard at localhost:3377.

$nexus-prime start

› Questions · [FAQ]

What you'd ask
on a discovery call.

What is Nexus Prime, in one sentence?

The agent runtime stack — five blocks (Memory, Routing, Sandbox, Bridge, Audit) that snap onto Claude Code, Cursor, Codex, and Windsurf. Your agents remember, stay in bounds, and cost less. One install for your whole stack.

Does it work offline?

Yes. Memory and routing run entirely on the developer’s machine. The only thing that needs the internet is your model gateway — and you choose what it is (Anthropic, OpenAI, local, or your own).

Which agents are supported?

12+ today. The big four — Claude Code, Cursor, Codex, Windsurf — plus Cline, Aider, Antigravity, Gemini, and more. One command per client: nexus-prime setup <agent-name>. Full list on Integrations.

How does it compare to Mem0, Zep, or LangGraph?

Mem0 and Zep are cloud memory APIs — Nexus runs on your box, no network needed for recall. LangGraph is a framework for building agents — Nexus is the runtime stack the agents run inside. The cost chart above draws from numbers each project published in their own paper or research page.

Does anything leave my machine?

By default, no. Memory and the dashboard are local. Only your chosen model gateway (Anthropic, OpenAI, local Ollama, etc.) sees prompts, and that gateway is configured by you — Nexus does not proxy through any third party.

Coding agents that finally ship.