trueforge

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

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.

BountyDesk | Bugs, CVEs, bounties. Reproduced securely.

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

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

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.

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.

Autonomous GitHub issue resolution agent that reproduces bugs in isolated sandboxes, generates and verifies minimal patches, and requires human approval before creating branches, pushing changes, or opening pull requests.

🧵 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.

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

On-call AI agent for Kafka dead-letter queues: safe, human-approved replays with zero double charges. Built on TrueForge.

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

TrueFoundry Hackathon '26: Homeward turns patient discharge paperwork into clear, source-linked next steps, with an AI assistant, human review, and care-team follow-up powered by TrueForge.

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.

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

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.

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.

Migration Rehearsal: an agent on TrueForge that rehearses every database migration PR on a masked copy of prod, fixes what breaks, and applies only through a human-gated, server-verified MCP tool.

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.

UPI payments ledger (Node + Postgres) with a live ops dashboard. Every migration PR is rehearsed on a masked copy of prod by Migration Rehearsal on TrueForge before it can merge.