Picado Labs

Deterministic, verify-first software systems for autonomous AI agents.

Open-source infrastructure for planning, routing, executing, verifying, and evaluating AI-driven software development.

Ecosystem • Projects • Quickstart • Repositories • Community

---

The Engineering Lifecycle

Picado Labs structures autonomous software development into disciplined, closed-loop stages:

$$\textbf{PLAN} \longrightarrow \textbf{ROUTE} \longrightarrow \textbf{EXECUTE} \longrightarrow \textbf{VERIFY} \longrightarrow \textbf{EVALUATE}$$

[ Software Goal / Spec ]
           │
           ▼
┌────────────────────────────────────────────────────────┐
│ 01 • BUILD-WITH-AI                                     │
│ PLAN  ──>  Context • Memory • Prompt Architect         │
└──────────────────────────┬─────────────────────────────┘
                           │
                           ▼
┌────────────────────────────────────────────────────────┐
│ 02 • MODEL ROUTER                                      │
│ ROUTE ──>  Sub-3ms Pareto (Quality, Cost, Latency)     │
└──────────────────────────┬─────────────────────────────┘
                           │
                           ▼
┌────────────────────────────────────────────────────────┐
│ 03 • TERMINAL AGENT                                    │ ◄────────────────────────┐
│ EXECUTE ──> psutil Sandbox • 12 Tools • Checkpoints    │                          │
└──────────────────────────┬─────────────────────────────┘                          │
                           │                                                        │
                           ▼                                                        │
┌────────────────────────────────────────────────────────┐                          │
│ 04 • INDEPENDENT VERIFIER                              │                          │
│ VERIFY ──> Automated Tests • AST Proof of Done         │                          │
└──────────────────────────┬─────────────────────────────┘                          │
                           │                                                        │
                 ┌─────────┴─────────┐                                              │
                 │                   │                                              │
               [PASS]              [FAIL] ─── Auto-Repair / Rollback Loop ──────────┘
                 │
                 ▼
┌────────────────────────────────────────────────────────┐
│ 05 • AGENTBENCH                                        │
│ EVALUATE ──> 12 Tasks • 5D SWE Score • Telemetry       │
└──────────────────────────┬─────────────────────────────┘
                           │
                           ├─ "Performance Feedback" ──> [ 02 • Model Router ]
                           └─ "Failure Insights"     ──> [ 03 • Terminal Agent ]

---

Core Projects

Project	Stage	Role	Tech Stack	Repository	Issue Tracker
build-with-ai	01 • PLAN	Zero-API CLI orchestrator & prompt architect	Node.js (>=16), TypeScript	GitHub • npm	Issues
Model Router	02 • ROUTE	Sub-3ms Pareto AI Traffic Control Room	Python (>=3.10), FastAPI, React 18	ai-model-router	Issues
Terminal Agent	03 • EXECUTE & VERIFY	Sandboxed verify-first autonomous coding agent	Python (>=3.10), Typer, psutil	terminal-agent	Issues
AgentBench	04 • EVALUATE	12-task benchmark suite & 5D SWE scoring platform	Python (>=3.10), Node.js (>=18), FastAPI	agent-bench	Issues

---

Projects Overview

1. build-with-ai - Plan & Orchestrate

• Zero-API, 100% Private: $0 cost, 0 API keys, local .buildwithai/ context store (state.json, context.json, history/).
• Dynamic Context Injection: Automatically injects architectural decisions into downstream prompts via {{decisions.key}}.
• 7 Production Workflows: Built-in templates for SaaS MVP, Full-Stack Web App, REST API, Mobile App, Flutter App, Chrome Extension, and AI Agent & RAG.

npx build-with-ai init
npx build-with-ai next

2. Model Router - Route & Optimize

• Sub-3ms Pareto Engine: Regex & heuristic analysis (<2.5ms) + in-memory multi-criteria scoring (<0.5ms) across Quality, Cost, Speed, Capabilities, and Reliability.
• Budget Control & Fallback: Tiered fallback (HTTP 429/503 retry cascading) and automated spend thresholds (80%, 95% local-only Ollama, 100% block).
• Traffic Control Room: FastAPI backend (port 8000) + React 18 dashboard (port 5173) with real-time SSE topology streaming and telemetry export.

python backend/app/cli/main.py doctor
python backend/app/cli/main.py route "Build async rate limiter in Python"

3. Terminal Agent - Execute & Verify

• Verify-First Architecture: Requires 0 test failures executed inside an isolated sandbox before declaring completion.
• Deterministic Context Engine: BM25 keyword + AST symbol ranking (def, class, function) without heavy vector databases.
• Process Sandboxing & Self-Healing: Local psutil process-tree termination, Docker sandbox, and 12-category automated error recovery.

terminal-agent doctor
terminal-agent run "Implement JWT auth middleware with pytest tests"

4. AgentBench - Benchmark & Evaluate

• 12 Standardized Tasks: Polyglot evaluation across Python (pytest) and Node.js (node --test).
• 5D Composite SWE Scoring: $0.50 \times \text{Correctness} + 0.25 \times \text{PassRate} + 0.10 \times \text{Quality} + 0.10 \times \text{Efficiency} + 0.05 \times \text{Repair}$.
• PR Ingestion & Live Dashboard: Ingests real GitHub PR diffs into executable tasks with React leaderboard and CI/CD gatekeeper.

python agentbench.py doctor
python agentbench.py run --benchmark fix-rate-limiter --provider ollama --model qwen2.5-coder:1.5b

---

5-Minute Ecosystem Quickstart

# 1. Initialize project with disciplined prompt architecture
npx build-with-ai init

# 2. Check toolchain readiness
python backend/app/cli/main.py doctor  # Model Router
terminal-agent doctor                 # Terminal Agent
python agentbench.py doctor           # AgentBench

# 3. Plan -> Route -> Execute -> Verify -> Evaluate
npx build-with-ai next
python backend/app/cli/main.py route "Build a secure token bucket rate limiter in Python"
terminal-agent run "Implement token bucket rate limiter with pytest tests"
python agentbench.py run --benchmark fix-rate-limiter --provider ollama --model qwen2.5-coder:1.5b

---

Repositories & Issues

• build-with-ai: PicadoLabs/build-with-ai • npm package • Issues
• Model Router: PicadoLabs/ai-model-router • Issues
• Terminal Agent: PicadoLabs/terminal-agent • Issues
• AgentBench: PicadoLabs/agent-bench • Issues
• Organization: github.com/PicadoLabs

---

Community & Maintainers

• Founder & Maintainer: Vardhman (Kap10) - GitHub • X (Twitter)
• Official Organization X: @PicadoLabs
• LinkedIn: Picado Labs
• Official Website: picadolabs.me
• Contact & Inquiries: picadolabs@gmail.com

---

_{Picado Labs • Open source. Practical AI infrastructure. Built to be verified.}