Foundry machinery

AgentFoundry is built from rebranded agent stack machinery.

The stack is not a menu of vendors. AgentFoundry absorbs the best agent apps, runtimes, UI layers, sandboxes, workflow frameworks, and eval systems into opinionated forge lines for building production-ready AI agents.

13agent-facing apps/* repos inventoried from lexopedia-stack
78agent/runtime/UI/sandbox vendor/* repos inventoried from lexopedia-stack
67agent/eval/runtime/training vendor/* repos inventoried from deepbrainz-stack
Rebrand, do not resell

Every upstream repo becomes AgentFoundry machinery before it reaches the customer.

The product promise stays simple: bring a workflow, leave with a forged agent. Internally, AgentFoundry can draw from apps/* and vendor/* to create custom blueprints, adapters, checks, interfaces, and deployment packs.

Opinionated — each repo maps to a forge line with a clear AgentFoundry job.
Rebranded — customer-facing language is AgentFoundry: blueprints, forge lines, quality lab, execution chamber, deployment handoff.
Composable — runtimes, tools, UI, sandboxes, workflows, and evals combine into one deployable agent system.
Forge lines

Blueprint Forge

Turns proven app patterns into AgentFoundry-owned templates for intake, configuration, run review, and deployment handoff.

Stack sources
  • apps/lexopedia-agent-builder
  • apps/vibe-coding-ai
  • apps/replicate-ai
  • apps/coding-agent
  • vendor/open-agent-builder
  • vendor/coding-agent-template
  • vendor/workflow-builder-template
AgentFoundry output: rebranded agent blueprints for coding, research, browser, document, support, and internal-workflow agents.

Runtime Forge

Normalizes agent runners into one opinionated AgentFoundry runtime contract instead of making customers choose vendors.

Stack sources
  • vendor/hermes-agent
  • vendor/openclaw
  • vendor/opencode
  • vendor/openai-codex
  • vendor/claude-code
  • vendor/gemini-cli
  • vendor/gitagent
  • vendor/deepagents
AgentFoundry output: planner, coder, browser, researcher, evaluator, reviewer, and handoff-agent adapters.

Workflow Forge

Composes specialist agents into supervised workforces with state, routing, approvals, retries, and human checkpoints.

Stack sources
  • vendor/langgraph
  • vendor/crewai
  • vendor/mastra
  • vendor/agno
  • vendor/pydantic-ai
  • vendor/openai-agents-python
  • vendor/openai-agents-js
  • vendor/microsoft-agent-framework
  • vendor/a2a
AgentFoundry output: AgentFoundry workflow graphs for single-agent, multi-agent, and human-in-loop business processes.

ToolMesh Forge

Converts connector ecosystems into governed tool access with typed contracts, permissions, approvals, and audit trails.

Stack sources
  • vendor/mcp-servers
  • vendor/mcp-python-sdk
  • vendor/mcp-typescript-sdk
  • vendor/mcp-inspector
  • vendor/mcp-registry
  • vendor/agent-skills
  • vendor/openai-skills
  • vendor/openclaw-skills
  • vendor/awesome-mcp-servers
AgentFoundry output: approved tools, skill packs, source access, API actions, and reusable integration recipes.

Interface Forge

Rebrands agent UI protocols and app shells into the AgentFoundry cockpit for configuring, supervising, and reviewing work.

Stack sources
  • vendor/vercel-ai
  • vendor/ag-ui
  • vendor/a2ui
  • vendor/copilotkit
  • vendor/microsoft/magentic-ui
  • vendor/vscode-copilot-chat
  • vendor/nesquena/hermes-webui
AgentFoundry output: live run streams, approval controls, trace viewers, lead intake, blueprint forms, and deploy previews.

Execution Chamber

Provides bounded sandboxes for code, browser, desktop, and computer-use workflows so agents can do real work safely.

Stack sources
  • vendor/e2b
  • vendor/e2b-code-interpreter
  • vendor/e2b-desktop
  • vendor/e2b-surf
  • vendor/e2b-open-computer-use
  • vendor/playwright
  • vendor/browser-use
  • vendor/browseros
  • vendor/browserless
  • vendor/lightpanda-browser
AgentFoundry output: sandbox receipts, screenshots, generated artifacts, command logs, browser traces, and rollback notes.

Engineering Workforce Forge

Uses the first-party engineering apps as the opinionated forge line for repo-connected agents and governed delivery workflows.

Stack sources
  • apps/gitclaw
  • apps/gitpilot
  • apps/gitwork
  • apps/gitcode
  • apps/deepcode
  • apps/deepcode-runtime
  • apps/deepagent
  • apps/deepwork-execution-cell
AgentFoundry output: repo-agent pilots for investigate → patch → verify → review → Proof Packet → PR-ready handoff.

Quality Lab

Hardens agents with evals, guardrails, security review, regression checks, tool-call verification, and evidence bundles.

Stack sources
  • deepbrainz-stack/vendor/evals/lm-evaluation-harness
  • deepbrainz-stack/vendor/evals/opencompass
  • deepbrainz-stack/vendor/evals/lighteval
  • deepbrainz-stack/vendor/evals/simple-evals
  • deepbrainz-stack/vendor/evals/frontier-evals
  • deepbrainz-stack/vendor/evals/verifiers
  • deepbrainz-stack/vendor/runtimes/guardrails
  • vendor/claude-code-security-review
  • vendor/openai-testing-agent-demo
AgentFoundry output: quality scores, risk registers, eval reports, policy checks, rollback paths, and foundry proof marks.

Industrial Plant

Keeps DeepBrainz training, inference, and ML substrate available for later scale without broadening the first pilot surface.

Stack sources
  • deepbrainz-stack/vendor/inference/vllm
  • deepbrainz-stack/vendor/inference/sglang
  • deepbrainz-stack/vendor/inference/llama.cpp
  • deepbrainz-stack/vendor/training/trl
  • deepbrainz-stack/vendor/training/axolotl
  • deepbrainz-stack/vendor/training/unsloth
  • deepbrainz-stack/vendor/frameworks/transformers
  • deepbrainz-stack/vendor/frameworks/pytorch
AgentFoundry output: future model serving, fine-tuning, evaluation, and cost/performance improvements for forged agents.
Build sequence
01Select

Choose one painful workflow and the closest AgentFoundry blueprint.

02Rebrand

Translate upstream app/vendor patterns into AgentFoundry-owned copy, schemas, UI, and contracts.

03Compose

Attach runtime, tools, workflow graph, UI shell, sandbox, approvals, and evidence capture.

04Harden

Run evals, policy checks, sandbox trials, traces, failure review, and human gates.

05Ship

Deploy as app, API, worker, repo workflow, browser workflow, or internal tool with a reviewable handoff.

06Improve

Turn traces, failures, evals, and customer feedback into better blueprints and forge-line defaults.

Anti-sprawl rule

Leverage everything. Expose only what helps one agent get forged.

This page documents the internal machinery. The landing page should still sell the business benefit: less repetitive work, faster handoffs, safer automation, deployable agents, and reviewable evidence.

Use stack breadth as foundry machinery, not as a public repo catalogue.
Fork, wrap, or adapt patterns into AgentFoundry-owned product surfaces before exposing them.
Keep each pilot anchored to one workflow, one owner, one success metric, and one deployment path.
Treat Proof Packets as quality certificates inside the foundry, not as the whole product identity.