Meshint · AI work estimator

Estimate AI work cost — no account needed

Describe a task and get a routing plan with cloud and local cost comparison instantly.

No card requiredLocal-first processingCloud vs local compared

Role distribution

Enter a task and start routing. The role distribution will appear here.

Cost estimate

The estimate will show role-level cost and local alternatives here.

Routing path

작업 입력 전 기본 경로
MODEL BROKER / AGENT RUNTIME / EVIDENCE MESH

Route Intelligence
Into Execution.

Meshint connects models, agents, workflows, and evidence into one operational mesh.
Break out of locked tool chains and turn requests into executable routing paths.

5 rolesRouter to Reviewer
mock/livebroker mode ready
evidenceprompt plan output
  • Model Broker selects models by task type and difficulty
  • Agent Runtime splits work into Router, Planner, Architect, Builder, and Reviewer roles
  • Insight Pipeline returns cost, time, and risk checks together
  • Evidence Mesh keeps envelope ids and execution prompts attached
Routing Score
78/100

Ready for dispatch

Latency p95
128ms-8.3%
Estimate
$0.047per 1k req
Evidence
Documents12,342
Datasets3,876
Signals8,125
Policies542
Model Broker
Llama 3.2 1B (iPhone tier)128ms
Llama 3.2 3B (iPhone tier)121ms
Qwen3 1.7B (iPhone tier)114ms
Gemma 3 1B (iPhone tier)107ms
Qwen3 4B (edge upper)100ms
Routing DecisionAuto

Awaiting input

Insight Pipeline
Ingest12.4K
Analyze9.7K
Synthesize2.1K
Act1.3K
Live Routing
Agent Runtime
All Agents 28Running 3Idle 6
Intent classificationqueued
Execution slicequeued
Proof summaryqueued
Proof Trace
classifyok
estimateok
dispatch-previewok

Browser device profile

Meshint profiles the browser in real time and shows the routing tier, compute headroom, and rendering limits before a request runs.

Not available yet

Not available yet

Signal trace

Not available yet

Compatible local models

Not available yet

Llama 3.2 1B (iPhone tier)

Locked

browser-webgpu · browser-wasm · ollama-local
Gemma 3 1B (iPhone tier)

Locked

browser-webgpu · browser-wasm · ollama-local
Qwen3 1.7B (iPhone tier)

Locked

browser-webgpu · ollama-local
Llama 3.2 3B (iPhone tier)

Locked

browser-webgpu · ollama-local
Qwen3 4B (edge upper)

Locked

browser-webgpu · ollama-local
Gemma 3 4B (edge upper)

Locked

browser-webgpu · ollama-local
Meta Llama 3.1 8B Instruct

Locked

ollama-local · self-hosted
Google Gemma 3 12B IT

Locked

ollama-local · self-hosted
Qwen3 14B

Locked

ollama-local · self-hosted
Mistral Small 3.1

Locked

self-hosted · cloud-api

Model efficiency by role

#RoleModelSuccess rateAvg. tokens
1BuilderGPT-5 mini94%1,200
2RouterGemini 2.5 Flash97%320
3ReviewerClaude Sonnet91%2,100
4PlannerGPT-5 mini88%900
5ArchitectClaude Sonnet85%3,400

* Live measurements are being collected. Current values are seeded examples.

© 2026 Meshint