trueforge

An agent that writes, verifies and publishes GPU kernels - and cannot lie about the speedup. Built on the TrueForge agent harness.

Build durable, governed, vendor-neutral agent harnesses with a visual Next.js studio, replayable runs, runtime approvals, and portable manifests.

Nothing reaches production without passing through the airlock. A change-control console where every irreversible production change is proven on a shadow copy in a sandbox before a human is ever asked to approve it. Built on TrueForge.

An access governance agent. Finds every grant your people and your AI agents still hold, waits for human approval, and writes a verifiable audit trail. Built on TrueForge.

An AI harness for video editing. Describe the edit; an agent makes it on your real timeline through 16 MCP tools: remove silences, caption every clip with parallel sub-agents, cut to the beat, or write the ffmpeg filter graph itself. It stops for your approval before anything destructive. Built on TrueForge, runs on any model provider.

PR reviewer that runs the PR in an agent-harness sandbox, reports what happened, and blocks only with human approval. Built on TrueForge.

Observability for TrueForge agents: trace viewer plus an investigator agent. Agent Harness Hackathon entry.

BountyDesk | Bugs, CVEs, bounties. Reproduced securely.

Replay-verified MCP security testing and deterministic policy hardening for tool-using AI agents

A TrueForge-powered, approval-gated agent that turns session-replay evidence into tested draft pull requests.

Incident response + analytics agent, built on the TrueForge harness. Investigates alerts with read-only tools, computes root cause in a sandbox, and stops for human approval before doing anything irreversible.

Evidence-grounded mission rehearsal for a PlayStation-teleoperated Unitree G1 in an Everest-inspired MuJoCo environment, with TrueForge approvals and Bright Data web research.

An agent that decides whether an npm package is safe to let into your repo and can't add it without your approval. Built on TrueForge.

AgentGuard is an assurance layer for AI agents. TrueForge runs the agent; AgentGuard verifies evidence, enforces policy, handles failure and recovery, and produces a deterministic assurance verdict with a live console.

Autonomous DevOps incident-response agent using TrueForge, MCP tools, and human-in-the-loop approval gates

🧵 Self-healing web data pipelines — an agent that detects scraper breakage, synthesizes validated selector repairs, and ships them only through a human approval gate. Built on TrueForge + Bright Data.

Evidence-gated runbook executor for incident response — an agent gathers evidence freely, but every production-changing action stays locked behind a human approval gate. Built on TrueForge (MCP) + Cloudflare Workers.

Learns with you at one hackathon. Builds with you at the next.

Autonomous operational reporting agent built on TrueForge with custom Slack MCP (SSE), sandboxed Python analytics, and human-in-the-loop safety gates.

A persistent self-correcting agent that learns from anomalies without self-issuing authority.