AISmush — The Intelligence Layer for AI Coding Agents

Game Changer

AI-Generated Project Agents

One command scans your codebase, sends it to AI for deep analysis, and generates agents customized to YOUR project — your patterns, your frameworks, your architecture.

Not generic templates. Agents that know your specific file structure, your naming conventions, your test framework, your build commands.

Scans your codebase in seconds
5-7 AI calls for deep analysis (~$0.03)
Generates agents, skills, and CLAUDE.md
Each agent assigned the cheapest model that can do the job
Resumes where it left off if interrupted
Incremental re-scans — only re-analyzes changed files (~$0.003)

$ aismush --scan

# Analyzing your codebase...
Detected: Rust + TypeScript + React
Type: fullstack web app (complex)

# Generating project-specific agents:
├─ rust-expert (sonnet) ✓
├─ frontend-engineer (sonnet) ✓
├─ test-runner (haiku) ✓
├─ debugger (sonnet) ✓
└─ explorer (haiku) ✓

Created 5 agents, 8 skills, CLAUDE.md

Included Free

24 Claude Code on Steroids Skills — Auto-Installed

aismush --scan auto-installs 24 best-in-class engineering workflow skills from Claude Code on Steroids alongside your project-specific agents.

These skills encode a complete engineering decision system — ORACLE classifies every task and routes to the right skill automatically. HUNTER hunts down bugs. FORGE builds features. ARCHITECT designs systems. The full cycle, optimized.

ORACLE — master routing skill, picks the right approach for any task
HUNTER / FORGE / GRADIENT — deep debug, code gen, safe refactoring
ARCHITECT / CHRONICLE / PATHFINDER — design, docs, exploration
MIT licensed — auto-cached, auto-updates on rescan
Replaces the previous 21-skill set with a more powerful, unified system
Combined with your project agents = a complete engineering team in one command

$ aismush --scan

# Project agents generated...

# Installing Claude Code on Steroids:
GadaaLabs/claude-code-on-steroids (MIT)

├─ ORACLE · HUNTER · FORGE
├─ CHRONICLE · ARCHITECT · GRADIENT
├─ NEXUS · PATHFINDER · VECTOR
└─ + 15 more covering the full dev lifecycle

24 skills installed. Cached locally.

Done. Project agents + CCoS skills ready.

Core Feature

Smart Model Routing + Blast-Radius Analysis

AISmush automatically detects what kind of work each turn requires and routes it to the cheapest model that can handle it.

Planning and architecture? Claude ($15/M). Reading files and making edits? DeepSeek ($0.27/M). That's a 55x cost difference on the turns that matter most.

Zero latency overhead — pure heuristic routing
Claude for reasoning, DeepSeek for execution
Blast-radius aware — parses imports to know which files are critical
Editing a shared type? Claude. Editing a leaf file? DeepSeek.
Automatic failover between providers
Error recovery detection (3+ errors → Claude)

# Works the same for any agent:

Claude Code or Goose or any client
↓
AISmush proxy
↓

# Routing decisions per turn:
"Plan the auth system" → Claude ($0.45)
Tool result: Read file → DeepSeek ($0.001)
Tool result: Edit file → DeepSeek ($0.001)
Tool result: Run tests → DeepSeek ($0.001)
"Debug this error" → Claude ($0.12)

Session: $0.58 instead of $12.40

New in v1.3.0

Native Code Graph — Symbol-Level Intelligence

AISmush builds a full AST-based code graph during --scan, using Tree-sitter compiled directly into the binary. No external tools. No Node.js. Pure Rust.

The graph knows every function, every struct, every call reference — and which symbols are depended on by the most code. Smarter routing. Automatic context. Visual exploration.

Extracts functions, structs, classes, traits, types, constants from 6 languages
Tracks cross-file call references — real symbol-level edges, not just imports
Symbol blast radius: BFS over call graph to score how critical each symbol is
Context auto-injected when you touch high-impact files (transparent, no config)
Incremental — only re-parses changed files on subsequent scans
Graph tab in dashboard: symbol search + blast radius heat map

$ aismush --scan .

# Building code graph (AST, Tree-sitter):
Parsed: 247 files in 0.8s
Symbols: 1,842 extracted
Refs: 4,391 call edges

# Blast radius computed:
src/db.rs → 0.91 (critical)
src/router.rs → 0.67 (high)
src/compress.rs → 0.23 (moderate)

# Auto-injected context on next edit of db.rs:
[graph] src/db.rs blast_radius=0.91
exports: open, log_request, get_stats

Biggest Token Saver

Structural Summarization — 3-5x Fewer Tokens

The biggest single improvement to token usage. Older tool results in your conversation get replaced with compact structural summaries — just function signatures, type definitions, and imports.

Your last 4 messages stay fully intact. Only older code results get summarized. JSON, YAML, and error results are never touched.

200-line file becomes ~30 lines (3-5x reduction)
Saves thousands of tokens per request in long sessions
Content-type aware — only summarizes code, never data
Supports Rust, TypeScript, Python, Go, and more
Combined with standard compression: 60-80% total reduction

# Old message tool_result (6,000 tokens):
use std::collections::HashMap;
use crate::db::Db;
// ... 180 lines of implementation ...
// comments, function bodies, tests ...

# After structural summary (1,200 tokens):
[Structural summary (200 lines -> 28 lines)]
use std::collections::HashMap;
use crate::db::Db;
pub struct ProxyState { ... }
impl ProxyState { ... }
pub async fn handle() -> Response { ... }
fn compress_text() -> String { ... }

5x reduction. API surface preserved.

Nobody Has Solved This

Deep Memory — Full Conversation Capture

Every developer's frustration: "I already told you this yesterday."

Other tools remember tool names. AISmush captures entire conversations — your questions, the AI's answers, the reasoning, the decisions. Searchable by meaning, not just keywords.

Full conversation capture (not just tool names)
Keyword search always on (FTS5, zero overhead)
Semantic search opt-in (--embeddings) — finds "JWT validation" when you search "auth bug"
Search via dashboard, CLI, or API
Auto-injects relevant past context into new sessions
Lightweight by default — no 90MB model unless you want it

$ aismush --search "auth token bug"

# Found 3 relevant conversations:

1. [3 days ago] score: 0.92
  You: "Fix the JWT token expiry bug"
  AI: "The issue is in validate_token().
      The expiry check uses > instead of >=..."
  Tools: Read(jwt.rs), Edit(jwt.rs), Bash(cargo test)

2. [5 days ago] score: 0.87
  You: "Add refresh token support"
  AI: "I'll create a refresh endpoint that..."
  - Added WebSocket broadcast server

Reliability

Context Window Management

Claude handles 200K tokens. DeepSeek handles 64K. Long sessions blow past DeepSeek's limit, causing failures and lost work.

AISmush automatically manages the mismatch. Old tool results get trimmed, large contexts route to Claude, and your work is never blocked.

Under 55K: both providers work fine
55-64K: trim old tool results for DeepSeek
Over 64K: auto-route to Claude (200K window)
Never breaks tool_use/tool_result pairing

# Context growing during long session:

Turn 1: 5K tokens → DeepSeek ✓
Turn 10: 25K tokens → DeepSeek ✓
Turn 20: 48K tokens → DeepSeek ✓
Turn 25: 58K tokens → compress + DeepSeek ✓
Turn 30: 72K tokens → auto-route to Claude ✓

# Without AISmush: DeepSeek fails at 64K.
# With AISmush: seamless handoff.

Transparency

Real-Time Cost Dashboard

See exactly what you're saving. Every request tracked: which provider, how many tokens, what it cost, what it would have cost at full frontier-model pricing.

Live dashboard at localhost:1849/dashboard
Per-request cost breakdown
Savings percentage with all-Claude comparison
Request history with full detail
Memory viewer
Stats persist across sessions in SQLite

Session Stats

Requests: 142
Claude turns: 12 (planning/debugging)
DeepSeek turns: 130 (execution)

Actual cost: $1.82
All-Claude cost: $18.40
Saved: $16.58 (90.1%)

Autonomous

Plan Orchestrator — Say "Go" and Walk Away

Ask your agent to make a plan, then say "run plan". AISmush analyzes every step, maps each one to the best specialized agent, figures out what can run in parallel, and executes the entire thing autonomously.

Reads plans agents already generate — no custom format needed
Maps steps to specialized agents (rust-expert, data-engineer, etc.)
Independent steps run in parallel for maximum speed
Context from completed steps feeds forward automatically
Verifies results with cargo check/test after completion
Always asks for confirmation before executing

You: make a plan to add auth

Agent writes plan with 5 steps...

You: go

PLAN: Add authentication (5 steps)
Wave 1: Step 1 → rust-expert
Step 2 → data-engineer
Wave 2: Step 3 → backend-engineer
Wave 3: Step 4,5 → test-runner

Ready to execute? [Go / No]

New in v1.1.5

Proxy Pool — Rotate IPs, Kill 429s

Claude rate-limits by IP address. One server, one IP, many users = 429 errors. The proxy pool distributes every Claude request across a list of outbound proxies in round-robin order — so no single IP ever absorbs the full load.

Combine with AISMUSH_MAX_CONCURRENT to throttle burst concurrency and keep rate pressure low at the source.

HTTP, authenticated HTTP (host:port:user:pass), and SOCKS5
Round-robin — one proxy per request, automatically
Automatic fallback to direct connection on proxy failure
Zero changes to your agent or workflow
Works alongside DeepSeek fallback for double-layer 429 defense

# config.json
{
  "proxies": [
    "proxy1.host:8080",
    "proxy2.host:8080:user:pass",
    "socks5://proxy3.host:1080"
  ],
  "maxConcurrentClaude": 5
}

# Each Claude request rotates:
Request 1 → proxy1 (1.2.3.4)
Request 2 → proxy2 (5.6.7.8)
Request 3 → proxy3 (9.10.11.12)
Request 4 → proxy1 (1.2.3.4) ...

Distributed load = no rate limit walls

The Intelligence Layer
for AI Coding Agents

Works with every AI coding agent

Stability & Security Release

Every AI coding agent has the same cost problem.

Everything You Need to Code Faster, Smarter, Cheaper

AI-Generated Project Agents

24 Claude Code on Steroids Skills — Auto-Installed

Smart Model Routing + Blast-Radius Analysis

Native Code Graph — Symbol-Level Intelligence

Structural Summarization — 3-5x Fewer Tokens

Deep Memory — Full Conversation Capture

Context Window Management

Real-Time Cost Dashboard

Plan Orchestrator — Say "Go" and Walk Away

Proxy Pool — Rotate IPs, Kill 429s

Three Steps. That's It.

Install

Scan

Code

Works With Any Agent, Three Ways

Claude Code (Default)

Direct Mode (Any Agent, No DeepSeek)

Goose / Other Agents

Install in 10 Seconds

The Intelligence Layerfor AI Coding Agents

Works with every AI coding agent

Stability & Security Release

Every AI coding agent has the same cost problem.

Everything You Need to Code Faster, Smarter, Cheaper

AI-Generated Project Agents

24 Claude Code on Steroids Skills — Auto-Installed

Smart Model Routing + Blast-Radius Analysis

Native Code Graph — Symbol-Level Intelligence

Structural Summarization — 3-5x Fewer Tokens

Deep Memory — Full Conversation Capture

Context Window Management

Real-Time Cost Dashboard

Plan Orchestrator — Say "Go" and Walk Away

Proxy Pool — Rotate IPs, Kill 429s

Three Steps. That's It.

Install

Scan

Code

Works With Any Agent, Three Ways

Claude Code (Default)

Direct Mode (Any Agent, No DeepSeek)

Goose / Other Agents

Install in 10 Seconds

The Intelligence Layer
for AI Coding Agents