Live · In production since March 2026

The autonomous AI workforce for founders.

Delegate your work across coding, research, content, and operations. Agents execute with full context of your business, using the cheapest model that can handle the task, within scopes you control. Every action is reversible and auditable.

Start 14-day trial Self-host (MIT)

$199 Solo · $499 Pro · $149/seat Team · Enterprise custom. 14-day trial, cancel anytime.

MIT

Open-source engine

8 mo

Production use

0

Catastrophic failures

arXiv

Benchmark paper in submission

Security, Trust Ladder, compliance roadmap
The problem

40-60 hours a month lost to context-switching.

A solo founder running 2-5 products uses 5-8 AI tools daily. Claude.ai for strategy. Cursor for code. Notion AI for writing. ChatGPT for research. A custom script for deployment. Each tool has its own memory, interface, and prompting pattern. None share context.

Every switch costs 5-15 minutes of re-explaining. Over a month, that's 40-60 hours gone — not to productive work, not even to meetings. Just to re-establishing context in tools that have zero memory of yesterday's conversation.

Koovis Workforce solves that by remembering everything — and acting on it.

How it works

Six things no competitor combines.

Multi-Domain Agents

Coding, research, content, operations, strategy. 5 squads with shared context across all your projects. Not a coding-only tool; not a chat-only tool.

Cross-Project Context

Agents know your entire business — every project, decision, task, and constraint. No more re-explaining context to different tools.

Trust Ladder

L0 Propose → L1 Scoped → L2 Autonomous-within-scope → L3 Hands-off. Agents earn autonomy by proving reliable. You control what they can do.

Credit-Aware Routing

5 providers (Claude, GPT, Gemini, DeepSeek, Bedrock). Automatic routing picks the cheapest model that can handle the task. Cost drops 60-80% vs single-vendor tools.

Persistent Memory

SQLite WAL + FTS5 memory layer. Agents remember your projects, decisions, preferences, and patterns across sessions. Context compounds.

Full Audit Trail

Every agent action logged with reasoning trace, cost, diff, and approval status. Reversibility by default — agents archive instead of delete, draft instead of send.

Trust Architecture

Agents earn autonomy.

Every user starts at L0 (every action proposed, you approve each one). Agents progress to higher levels as they prove reliable. Sensitive actions (git push main, DB delete, external email, payments) require approval regardless of level.

L0

Propose

Every action is a proposal. You click approve or reject on each one.

Default for new users + new squads. Safe. Trust-building.

L1

Scoped

You pre-approve action categories (e.g., "this agent can write to `scratch/`, open PRs, post to `#agent-output`").

After 5 successful L0 merges for that squad.

L2

Autonomous-within-scope

Agent runs continuously within defined scope. You review end-of-session summary.

After 30 days at L1 + 100 successful runs.

L3

Hands-off (expert)

Full autonomy within hard guardrails. You review weekly digest.

Opt-in only, with explicit acceptance of risk.

Open-core

MIT engine + hosted SaaS.

Same pattern as Supabase, Cal.com, PostHog, Neon. The core engine is MIT-licensed and lives at github.com/koovis-ai/koovis-workforce. The hosted SaaS wraps the engine with multi-tenant auth, billing, onboarding, and team features.

Engine (MIT open-source)

  • Multi-provider chat backend (5 providers)
  • Credit-aware model router with fallback
  • Agent orchestrator (squads, budgets, worktrees)
  • 13 bundled MCP tools
  • Memory store (SQLite + FTS5)
  • Starter squad templates

Hosted SaaS (proprietary)

  • Multi-tenant auth + workspace isolation
  • Billing (Stripe + Razorpay, GST-compliant)
  • Guided onboarding + integration wizards
  • Approval workflow UI + audit dashboard
  • Team features (shared workspaces, admin)
  • Plugin marketplace (community tools)
Pricing

Priced between chat tools and hiring.

A Koovis agent doing 20-40 hours of work per month at $50-150/hr contractor rate = $1-6K of value. We capture a fraction of that. Positioned above chat tools ($20) and below hiring ($5K+/mo). Anchored against contractor replacement.

Self-hosted

Free

MIT open source

Devs, privacy-conscious, community

  • Full engine, yours to run
  • Bring your own provider keys
  • 13 bundled MCP tools
  • Community support (Discord)
View on GitHub

Founder Solo

$199

per month

Solo founders, indie hackers

  • 1 workspace, 3 squads
  • 500 agent-runs/month
  • Hosted, no ops work
  • Email support
Start 14-day trial
Primary

Founder Pro

$499

per month

Multi-product founders

  • 3 workspaces, all squads
  • 2,000 agent-runs/month
  • Cross-project orchestration
  • Priority model access
  • Priority support
Start 14-day trial

Team

$149

per seat/month (min 5)

AI-first teams, small agencies

  • Shared workspaces
  • Team memory + admin
  • Audit log export
  • Slack support channel
Contact sales

Enterprise: $1,500-5,000/seat/mo — SSO, SLA, audit retention, on-prem option, dedicated support. Contact sales.

Security

Built for the 3 AM scenario.

Reversibility by default

Agents archive instead of delete. Open PRs instead of pushing to main. Draft email instead of sending. The blast radius of any action is bounded.

Hardcoded approval gates

Sensitive actions require approval regardless of trust level. Git push main, DB delete, external email, payments — always gated.

Full audit trail

Every agent action logged with reasoning trace, cost, and diff. Query history, replay sessions, spot anomalies.

Self-hosted option

Full data sovereignty. Run on your own infra. No phone-home, no telemetry. Your workspace context stays with you.

Responsible disclosure: security@koovis.ai · SECURITY.md

FAQ

Common questions.

How is this different from Devin, Cursor, or Manus?+
Devin is coding-only at $500/mo. Cursor is synchronous editor-based. Manus had persistence issues. Koovis Workforce is multi-domain (coding + research + content + ops + strategy), has persistent cross-project memory, and uses a Trust Ladder so agents earn autonomy instead of getting it all upfront. Positioned between “chat tools” ($20/mo) and “hiring a contractor” ($5K+/mo) — aimed at multi-product founders.
What makes autonomous agents actually safe?+
Three architectural principles, hardcoded into the engine:
  1. Trust Ladder. Agents start at L0 (propose-only). Users earn higher autonomy after proven reliability.
  2. Reversibility by default. Agents archive instead of delete, draft instead of send, open PRs instead of pushing to main.
  3. Sensitive actions always approval-gated. Git push to main, DB deletes, external email, payment operations — require explicit approval regardless of ladder level.
Is the engine really open-source?+
Yes. The core engine — chat backend, multi-provider router, agent orchestrator, MCP tool registry, memory layer — is MIT-licensed at github.com/koovis-ai/koovis-workforce. Fork it, run it yourself, build on top. The hosted SaaS bundles product-layer features (multi-tenant auth, billing, onboarding, team features, plugin marketplace) — those are proprietary.
Can I self-host and skip the hosted SaaS?+
Absolutely. The OSS engine runs standalone — laptop, VM, Docker, your own cloud. Bring your own provider keys (Claude, OpenAI, Gemini, Bedrock, DeepSeek). No phone-home, no telemetry by default. Hosted SaaS exists for users who want someone else to handle ops.
What providers does credit-aware routing support?+
Five at v0.1: Claude (direct + AWS Bedrock), OpenAI, Google Gemini (direct + GCP Vertex AI), DeepSeek. The router reads your credit balances and picks the cheapest provider that can handle each task. Auto-fallback on auth/billing/rate-limit errors. More providers via config; community-contributed.
Do you train on my data?+
No. We don't train models on your conversations, workspace content, or agent actions. Each provider (Claude, OpenAI, etc.) operates under their own privacy terms. We pick enterprise-grade defaults where data is not retained for training. See our Privacy Policy for specifics.
What if agents break something?+
The Trust Ladder is designed to prevent this. L0 = every action is a proposal you approve. Reversibility by default means the blast radius of a bad action is a PR you reject, not production downtime. Sensitive actions (git push main, DB delete, email send) always require approval. In 8 months of in-house production use since March 2026, zero catastrophic agent failures.
What's the roadmap?+
  • Now: Closed Alpha (early-access pricing for the first cohort of founders)
  • Month 2: Private Beta at full pricing tiers
  • Month 3: Public launch with comparative agent-framework benchmark paper on arXiv
  • Month 4-6: Team features, plugin marketplace, regional vertical editions
  • Month 9-12: Enterprise tier (SSO, audit retention, on-prem), native mobile
See full plan on our blog.

Delegate. Don't chat.

Start a 14-day trial. Cancel anytime. Or self-host the MIT engine today.

Start 14-day trial Self-host