The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Local workspace for Claude Code, Codex, and Gemini CLI
Velocity is a local-first workspace that captures agent sessions, shows analytics, and helps you reuse repeated work. It connects console runs to sessions, review tools, routing views, and workflow builders so you can inspect what happened and turn it into something repeatable.
Builders who want one place to run Claude Code, Codex CLI, or Gemini CLI and turn sessions into reusable workflows.
You can inspect real agent work, spot what is worth keeping, and reuse it as workflows, agents, skills, or commands.
What it does
Session capture and review
Records local CLI sessions and gives you session detail views, transcript filters, scrolling, and compare workspaces.
Analytics and usage views
Shows usage, cost, latency, tools, and routing context so you can check whether agent work is behaving well.
Workflow builder
Lets you build and edit workflows from repeated work, including AI-assisted setup and node prompt editing.
Agents, skills, and commands
Provides surfaces for creating and managing reusable agents, skills, and commands at project or global scope.
Routing and model views
Includes routing graphs and model management so you can see entrypoints and provider behavior more clearly.
Hooks, MCP, and plugins
Supports hooks, MCP, plugins, and settings so the workspace can connect to tools and extend its behavior.
How to get it
- 1Regenerate all demos
python3 scripts/demo/capture-demos.py --base-url http://127.0.0.1:3000 --provider claude --target-seconds 12
- 2Run
git clone https://github.com/OptimiLabs/velocity.git cd velocity bun install bun dev
- 3If Console PTY launch fails on macOS during setup, run
chmod +x node_modules/node-pty/prebuilds/*/spawn-helper
- 4Reinstall with Bun
rm -rf node_modules bun install
- 5If PTY launch fails because spawn-helper is not executable
chmod +x node_modules/node-pty/prebuilds/darwin-arm64/spawn-helper # or (portable across prebuild folders) chmod +x node_modules/node-pty/prebuilds/*/spawn-helper
README
Velocity
Velocity is a local-first command center for Claude Code, Codex CLI, and Gemini CLI. It helps you run sessions, understand what happened, and turn repeated work into reliable workflows.
Why Velocity
Most CLI-only workflows break down as your team scales:
- session context is fragmented
- routing/entrypoint behavior is opaque
- reusable flows stay tribal knowledge
Velocity addresses this with three core loops:
- Run work in local CLI sessions.
- Inspect usage, cost, latency, tools, and routing context.
- Reuse what works as agents, workflows, skills, and commands.
Preview
These GIFs are embedded directly in the README and render on GitHub.
Workflow Builder
AI Assist setup -> finished workflow canvas -> AI prompt review -> node prompt editing.

Routing Graph
Fullscreen routing deep-dive with readable node zoom and multi-node inspection.

Sessions Journey
Console -> Sessions -> Session Detail walkthrough with transcript filters and scrolling.

Review Compare
Two-session review workspace with compare scope and analysis controls.

Back-to-Back Reel
All demos played continuously.

Stitched Reel
Continuous reel with short holds between each section.

Regenerate all demos:
python3 scripts/demo/capture-demos.py --base-url http://127.0.0.1:3000 --provider claude --target-seconds 12
Start in 60 Seconds
git clone https://github.com/OptimiLabs/velocity.git
cd velocity
bun install
bun dev
If Console PTY launch fails on macOS during setup, run:
chmod +x node_modules/node-pty/prebuilds/*/spawn-helper
Open http://localhost:3000.
Workflow-First Mental Model
If you only learn one thing, learn this:
- Use Console and Sessions to capture real work.
- Convert proven patterns into Agents and Workflows.
- Use Routing and Analytics to validate that automation is actually correct and cost-efficient.
Feature Inventory
Velocity is organized into three layers:
| Layer | Purpose | Key surfaces |
|---|---|---|
| Workspace | Run and inspect | Console, Sessions, Review, Analytics, Usage |
| Build | Reuse and automate | Agents, Workflows, Skills, Commands, Hooks, MCP, Routing |
| Platform | Configure and extend | Models, Plugins, Marketplace, Settings |
Detailed inventory: docs/guides/functionality-inventory.md
Provider Parity
| Capability | Claude | Codex | Gemini |
|---|---|---|---|
| Sessions + analytics ingestion | Yes | Yes | Yes |
| Agents (global + project scope) | Yes | Yes | Yes |
| Workflows | Yes | Yes | Yes |
| Skills + commands | Yes | Yes | Yes |
| Routing graph | Yes | Yes | Yes |
| Hooks | Yes | No | Yes |
| Plugins | Yes | No | No |
Core Tech
| Layer | Stack |
|---|---|
| Framework | Next.js 16 (App Router) |
| Runtime / PM | Bun |
| DB | SQLite (better-sqlite3) |
| State | TanStack Query + Zustand |
| Real-time | WebSocket + PTY manager |
| Graph UI | @xyflow/react |
| Tests | Vitest |
Docs
- Feature docs:
docs/features/ - Install + config guides:
docs/guides/ - Provider matrix:
docs/guides/provider-capability-matrix.md
Troubleshooting
better-sqlite3 binding errors
Reinstall with Bun:
rm -rf node_modules
bun install
Console terminal blank / stuck
- Verify
bun devstarts without PTY errors. - Hard refresh browser.
- Clear persisted local layout state.
node-pty spawn-helper permission denied (macOS)
If PTY launch fails because spawn-helper is not executable:
chmod +x node_modules/node-pty/prebuilds/darwin-arm64/spawn-helper
# or (portable across prebuild folders)
chmod +x node_modules/node-pty/prebuilds/*/spawn-helper
Then restart bun dev.
Security
Velocity is local-first. Data stays on your machine by default. Optional remote calls are limited to explicit user-driven actions (for example marketplace sources).
See SECURITY.md for reporting and policy details.
Inquiries
For questions or inquiries, contact: jaewonlee9642@gmail.com
Contributing
See CONTRIBUTING.md for setup and PR workflow.
Files in the repo
- __tests__
- .github
- app
- components
- docs
- hooks
- lib
- public
- scripts
- server
- stores
- types
- .gitignore
- bun.lock
- bunfig.toml
- CHANGELOG.md
- CLAUDE.md
- CODE_OF_CONDUCT.md
- components.json
- CONTRIBUTING.md
- eslint.config.mjs
- instrumentation.ts
- INTENT.md
- LICENSE
- next.config.ts
- package-lock.json
- package.json
- postcss.config.mjs
- README.md
- SECURITY.md
- tsconfig.json
- vitest.config.ts
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More harnesses
from vibe coding to agentic engineering - practice makes claude perfect
🌊 The original agent meta-harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
Practical patterns, starters & CLI tools for loop engineering with AI coding agents. Design systems that prompt and orchestrate agents (inspired by Addy Osmani and Boris Cherny). Includes loop-audit, loop-init, loop-cost.
Git. Ship. Done - Core

The most RAM efficient harness