LIVE SESSION ACTIVE

Code is
Conversation

AI and humans write code together in real time. Spawn agents, iterate in natural language, ship in minutes — not hours.

8
Specialist Agents
3
AI Models
∞
Conversations
nimbus-session — kimi-code@pair-42
$ nimbus spawn kimi-code --session=pair-42
> Spawning Kimi agent...
> Reading context: "implement binary search tree"
> Claude: Let me start with the node structure.
class Node {
constructor(val) {
this.val = val;
this.left = this.right = null;
}
}
> Kimi: Inserting BST logic...
> Tests passing: 12/12 ✓
> Session complete. 4 agents collaborated.
SCROLL
LIVE EDITOR

Write Once, Three Models Think

Every keystroke is analyzed by Kimi, Claude, and Gemini simultaneously. Suggestions appear inline — accept, reject, or remix.

pair-session.python
1# AI-assisted: merge sort implementation
2# Kimi suggested the divide step
3
4def merge_sort(arr):
5 if len(arr) <= 1:
6 return arr
7
8 mid = len(arr) // 2
9 left = merge_sort(arr[:mid]) # recurse left
10 right = merge_sort(arr[mid:]) # recurse right
11
12 return merge(left, right)
13
14def merge(left, right):
15 result = []
16 i = j = 0
17
18 while i < len(left) and j < len(right):
19 if left[i] <= right[j]:
20 result.append(left[i])
21 i += 1
22 else:
23 result.append(right[j])
24 j += 1
25
26 return result + left[i:] + right[j:]
27
28# Claude: Added edge case handling
29arr = [64, 34, 25, 12, 22, 11, 90]
30print(merge_sort(arr)) # [11, 12, 22, 25, 34, 64, 90]
31
AI SUGGESTIONS
K
KimiL3

Consider memoizing this with functools.lru_cache for repeated calls

C
ClaudeL18

The merge function allocates a new list — for large arrays, in-place would be O(1) space

G
GeminiL5

Add type hints: arr: list[int] → list[int] for better IDE support

3 agents active · 2ms latency
SESSION VISUALIZATION

Watch Agents Collaborate Live

Subagents spawn on demand, each with a focused role. Tasks flow through a FIFO queue — no bottlenecks, no wasted cycles.

Planner
Orchestrator
DONE

Decomposed task into 4 subtasks

via Claude100%
Coder
Implementation
RUNNING

Writing merge sort + tests

via Kimi72%
Reviewer
Code Review
THINKING

Analyzing time complexity

via Claude40%
Docs
Documentation
IDLE

Waiting for Coder

via Gemini0%
FIFO TASK QUEUE3 pending
1✓
Parse user intent
Planner · Claude
T-001
2✓
Generate BST skeleton
Coder · Kimi
T-002
3▶
Write merge sort
Coder · Kimi
T-003
4○
Review complexity
Reviewer · Claude
T-004
5○
Write docstrings
Docs · Gemini
T-005
6○
Run test suite
Coder · Kimi
T-006
MODEL SUPPORT

Three Models, One Conversation

Each model has a role. Kimi leads, Claude reviews, Gemini visualizes. The orchestrator routes tasks to the best model automatically.

K
Primary

Kimi

Built for code

Moonshot AI's frontier model optimized for long-context code generation and multi-file reasoning.

Long context (128k)Multi-file editsRepo-level understandingFast iteration
C
Core

Claude

Thinks before it codes

Anthropic's Claude brings careful reasoning, excellent documentation, and safe code generation.

Deep reasoningCode reviewDocs & testsSecurity analysis
G
Extended

Gemini

Multimodal context

Google Gemini for architecture diagrams, schema generation, and data pipeline design.

Diagram readingSchema designAPI surfaceData modeling
SUPPORTED LANGUAGES
🐍Python
📘TypeScript
🐹Go
🦀Rust
☕Java
🗃️SQL
🔧Shell
📝YAML
BUS MONITOR

Real-Time Task Bus

Every agent message flows through a central HTTP bus. Watch tasks arrive, get processed, and resolve — in real time.

LIVE
http://127.0.0.1:18785
♡02:14:15#sprint-1heartbeatAll agents alive · latency 2ms
✓02:14:08#sprint-1kimi-k1Draft complete: 28 lines, 2 functions
→02:14:02#sprint-1coderStarted: write_merge_sort task T-003
→02:14:01#sprint-1plannerDecomposed: implement merge sort → 4 tasks dispatched
messages5
channels4
agents4
errors0
API PLAYGROUND

Integrate in Minutes

REST + SSE API. Spawn sessions, stream tokens, chat with agents — all from a single endpoint. OpenAI-compatible schema.

Auth-ready — JWT middleware configured. Protected routes require Authorization: Bearer
POST/api/sessions
{
  "task": "Implement a binary search tree with insert, search, and delete",
  "models": [
    "kimi",
    "claude"
  ],
  "language": "python",
  "context": "production-ready, with type hints and tests"
}
Requires active session · Auth: Bearer <token>