Solopreneurs and small teams already pay $500–$2,000/month for Claude, Perplexity, Gemini, Codex, ChatGPT, and the rest. They use maybe 10% of it. Sandolab is the workbench that lets one operator actually leverage all of it — coherently, with budgets, audit, and handoff.
Not a reseller. Not another wrapper chat. A control plane that routes work across the seats you already own, pools accounts inside a small team, and turns idle subscription capacity into compounding research, code, and product output.
The market is convinced AI economics means token-metered API calls. That misses where the actual capacity is sitting: in flat-rate consumer and prosumer subscriptions that 99% of users barely touch.
Every research query, every coding agent, every browser scrape meters against an API key. Costs scale linearly with ambition. Solo founders cap themselves at $50/mo of API and never use the full power of the tools.
Claude Max, Perplexity Pro, Gemini Advanced, ChatGPT Pro, Codex are flat. They include orders-of-magnitude more capacity than any one human can manually consume. The marginal cost of an automated query is $0.
A typical operator already pays for Claude Max ($200), ChatGPT Pro ($200), Gemini Advanced ($20), Perplexity Pro ($20), GitHub Copilot ($20), Google AI Ultra ($250 — with $100 GCP credits attached). That's ~$700/month sitting on the desk. Used the way most people use it, it's a glorified search box. Used with Sandolab, it's a 24/7 research, coding, and synthesis fleet that would cost $5K–$20K/month to replicate on raw APIs.
We're not selling cheaper compute. We're selling the difference between "I have a Perplexity Pro tab open" and "Perplexity is running 100 structured research queries for me overnight, every night, feeding into a synthesis layer I read with my coffee."
Six layers, already running in production against my own work and in a small pilot pool.
Operator-held credentials for Claude, ChatGPT, Gemini, Perplexity, Codex, GitHub, and the rest. No reseller markup, no shared logins, no credential bleed across profiles.
A scheduler decides which task goes to which seat. Cheap reasoning to Haiku, deep research to Perplexity browser, codegen to Claude Code, synthesis to Gemini. Daily caps protect quota.
A Chrome extension operates the actual AI web UIs the operator already pays for. Programmatic queries against flat-rate plans. No API token billing.
Every conversation, every research deposit, every output ingests into a single Postgres-backed knowledge graph. Beliefs, contradictions, follow-ups, lineage.
A research flywheel reads the deposits, synthesizes, spawns follow-up tasks, updates beliefs, and surfaces what the operator should actually look at. The system gets sharper the longer it runs.
Operator dashboard shows live task queue, per-seat budget, fleet health, cost pressure, and a full audit trail. Friends in a small team get email-gated access with per-person caps.
Not a deck. The system is currently running my consulting work, research pipeline, and the first pilot pool of users.
The kernel: Go + Postgres mesh that holds task state, agent registry, research topics, beliefs, goals, cost ledger. Single source of truth across distributed runners.
Chrome extension that drives ChatGPT, Claude.ai, Gemini, Perplexity programmatically through their web UIs. The piece that turns flat-rate plans into automation.
LibreChat + LiteLLM stack with per-friend monthly budget caps. Friends sign in by email; never see credentials; spend mapped per-person. Subscriptions stay with the operator.
Spawns Claude Code, Codex, Gemini CLI runners in isolated worktrees. Routes tasks by capability and cost. Steerer pattern fans out cheap subagents from one parent.
Topics → deposits across Claude/Gemini/Perplexity in parallel → synthesis with agreement and disagreement → follow-up tasks → the loop tightens.
React + XState dashboard. 29 views across queue, dispatch, topology, research, problems, fleet health, terminal, semantic context.
Not enterprises. Not the "AI for everyone" crowd. The operator who's already running 6 tabs of AI tools and knows they're leaving 90% of the value on the table.
Building real product. Need a research, coding, and ops layer that scales without a team. Already pay for the tools.
Running multiple engagements. Need parallel research lanes per client, audit trails, and budget walls.
Don't want enterprise licenses or shared logins. Want one operator-held stack with per-person caps and clean handoff.
Scoping and pricing happen on a call once we've looked at your actual stack.
Customer owns every subscription, every credential, every cloud project. No reseller markup. No platform lock-in. We can't hold them hostage even if we wanted to.
18 months of fighting Chrome extension drift, selector breakage, vision-layer verification, profile isolation, and quota math. Hard to copy; even harder to keep running.
A purpose-built Go + Postgres orchestration system with a quorum model, durable workflows, and 100+ MCP tools. Three years of design decisions baked in.
Daily snapshots of plan changes across all major providers, with a human review queue. The routing layer knows about a new Claude tier before the customer does.
Every route decision, every dispatched task, every cost charge is logged. Customers can leave with their data, their repos, their accounts intact.
This is not a vaporware deck. The stack runs my own consulting work daily, dispatches research overnight, and serves a small pilot pool of paying users.
The stack works. What I need is the runway to harden it, productize the workbench, and bring the pooled-lab offering to a wider group of operators.
Polish the operator console, onboarding flows, and the chat-pool admin surfaces so a new customer can stand up the stack in a day.
Vision-layer verification across every supported provider, automated UI-drift detection, and self-healing selectors when the AI vendors ship UI changes.
Move from a handful of users to a meaningful cohort across solo-founder, consultant, and small-team profiles. Build the case-study layer.
If the subscription-leverage thesis lands for you and you back technical solo founders building real systems, I'd like to walk you through the live stack and the pipeline of pilot users.