parviSight sees the web. parviClaw-core executes commands. parviALMA remembers. Each powerful alone. Together, a force multiplier — all governed by parviGov.
FriendlyAI builds three agent capability tools. Each is powerful on its own. Together, they form a complete agent execution, observation, and memory platform, all governed by parviGov.
26 MCP tools. Agents that see the web, click, scroll, extract, fill forms, and interact with human-grade fidelity. Every action produces a cryptographic leaf.
Real PTY terminals. Not chat wrappers. Multiple LLMs through one governed pipeline. Coding plans, lifecycle management, quality gates.
Governed, persistent memory that learns from every engagement. Agents accumulate knowledge across sessions, decisions, and actions.
Purpose-built, not a repackaged testing framework. Dual-projection architecture: a human sees a browser window, an agent sees a structured accessibility tree with stable reference tags.
Not a chat box. A chat box cannot spawn a process. A chat box cannot kill a runaway agent. parviClaw-core is a Rust multi-session control plane with real PTY terminals, cryptographic leaves, and a durable ledger.
Coding plans: Structured plans define what an agent builds, step by step, with checkpoints and verification at each stage. Each step produces evidence.
Governed, persistent agent memory that learns from every engagement. What is remembered is as accountable as what is done.
parviALMA is a governed memory system designed for agent fleets.
The closed loop: agents act, actions are governed, governance produces proof, proof builds trust, trust enables deployment, deployment generates data, data feeds memory, memory makes agents smarter.
A QA team needs to test a web application end to end. Here is how the three tools work together.
Navigates flows, fills forms, clicks buttons, extracts results. Every action governed. Every page state captured with snapshot or screenshot. Every assertion backed by cryptographic evidence.
Spawns test sessions, runs setup scripts, manages test data, executes teardown. Coding plan defines the test sequence. Quality gates verify each step passed before advancing.
Which flows are flaky. Which selectors break. Which environments behave differently. Next time, agents already know what to watch for.
parviGov governs the entire engagement. When the test report is delivered, it comes with a cryptographic chain of custody. This is not a test runner. It is a governed, self-improving testing system.
A development team needs to build a feature. parviClaw-core spawns coding agents. parviSight verifies visually. parviALMA remembers the patterns.
Claude Code or Codex sessions launched through the governed pipeline. Each agent works in its own PTY terminal, writing code, running tests, committing changes. Coding plans define scope step by step.
Once implemented, parviSight agents open the application, navigate to the new functionality, and verify it works. Screenshots and snapshots prove what was tested. No more claiming done without proof.
What worked. What failed. Which approaches the team prefers. The next feature build starts smarter.
parviGov governs the entire cycle. Every code change is a governed act. parviGov Checkpoint writes .ack files after each step, so the orchestrator knows what was completed even after a crash.
A security team needs to assess an application's attack surface. parviSight scans. parviClaw-core exploits in sandbox. parviALMA remembers patterns.
Agents browse the target, inventory exposed endpoints, map the attack surface, identify vulnerabilities. Stealth and anti-detection let agents operate without triggering defenses. Network interception captures traffic.
Vulnerabilities verified in isolated, governed terminals. Exploits executed safely. Every command recorded as a cryptographic leaf. Damage contained. Evidence sealed.
Which vulnerabilities recur. Which exploitation techniques work. Which defenses were bypassed and how. The next assessment starts with accumulated knowledge.
parviGov governs every scan, probe, and exploit. When the security report is delivered, it comes with cryptographic proof of what was tested, what was found, and what was exploited. Not opinions. Evidence.
Each tool can be purchased standalone for teams that need one specific capability.
All three tools together form a complete agent platform. The bundle is priced as a platform, not three separate products.
Agents that see.
Agents that prove.
Real terminals, real execution,
real control.
Agents that remember.
Governed, compounding knowledge.
Three tools. Sold separately or together.
All governed by parviGov. All built in-house by FriendlyAI.
All designed for a world where AI must prove what it does. Choose one tool for a specific need, or bundle all three for the full closed loop.
parviSight, parviClaw-core, and parviALMA — available individually or as a governed platform.