Hi, I'm Jalil, Founding Engineer at OpenWork, based in San Francisco. I like taking capabilities that are powerful but hard to use and making them simple, honest about what they're doing, and reliable enough to trust with real work.
Alongside OpenWork, I build small open-source tools for people working with agents: configuration, MCP, testing, and evals. Before agents, I spent 8+ years shipping software for enterprises and startups. That includes BetterFormula, a Salesforce formula editor with 13k+ users.
OpenWork is the open-source alternative to Claude Cowork. It's a desktop app plus one MCP that let you reuse the same skills, MCP servers, and connected services across your agents, teammates, and machines. I've landed 580+ PRs there since July 2026. By commits, I'm its #2 contributor.
Merged PRs per week, as a table
| Week of | Merged PRs |
|---|---|
| Jul 6 | 5 |
| Jul 13 | 49 |
| Jul 20 | 57 |
| Jul 27 | 51 |
| Aug 3 | 35 |
| Aug 10 | 50 |
| Aug 17 | 20 |
| Aug 24 | 67 |
| Aug 31 | 64 |
| Sep 7 | 105 |
| Sep 14 | 71 |
Some of what I've built there:
| Area | Highlights |
|---|---|
| MCP Apps | Interactive MCP App UIs inline in conversations (#3700), served through a credential-bound gateway (#4004), and assembled into organization dashboards (#4056, #5006) |
| Connect and the MCP gateway | An enterprise MCP client with deep diagnostics (#2694, #2975), OAuth conformance (#2771), a managed local OAuth gateway (#3652), stateless MCP (#4055), and Client ID Metadata Document clients (#5293) |
| Automations | Cloud-scheduled automations that run on desktop runners (#3466), automations proposed straight from chat (#3552), web automations in OpenWork Cloud (#3716), and saved Code Mode scripts (#3656) |
| Reliability and speed | Live sessions that survive engine rollovers (#3696), interrupted runs you can resume (#4124), block-by-block streaming (#4422), and virtualized long transcripts (#5089) |
| Browser and computer use | A built-in browser for every conversation (#4434), site tools with user takeover (#4468), and scoped native computer-use sessions (#4463) |
| Plugins and skills | Portable Agent Plugins end to end (#3594), GitHub plugin import (#2568), and organization skills served over MCP (#2567, #5208) |
| Cloud runtime and connectors | A provider-neutral cloud runtime with Daytona as one adapter (#4445), working Gmail, Calendar, Drive, Sheets, and Microsoft 365 actions (#4693), and PDF attachments for every model (#4334) |
More from the workshop
| Project | What it is |
|---|---|
| sysone · System One Bench | Gives an agent Jev as code (typed decisions composed in one MCP call), plus a public record of what the evidence does and doesn't show |
| LiteMCP Composer | One governed MCP endpoint for every user and agent, composed from many MCP servers |
| skills-kit | A repo-local switchboard for agent skill libraries |
| MCP Apps gallery · Workplace MCP Apps | Hosted, forkable MCP Apps you can try by URL and copy to build your own |
| NottyDuck WebMCP Playground | A tiny local-first playground where a person and an agent play together through WebMCP |
| freestyle-volumes | Persistent volumes for Freestyle VMs, backed by any S3-compatible bucket |
| flashpod | A TypeScript-first wrapper for RunPod Flash |
| prettui | Composable TypeScript packages for exact-frame terminal UIs |
| web-seek | Browser QA briefs: demonstrate a flow in Chrome, then hand a validated brief to a QA agent |
| SenseSight | Turns robot sensor streams into 3D spatial memory you can explore in the browser |
| BetterFormula | An IDE-like editor for Salesforce formulas, shipped as a browser extension with 13k+ users |
I'm on the maintainer side of a busy open-source repo every day, so I try to send the kind of PR I'd want to receive:
- Small and focused. One problem per PR, with no drive-by refactors or formatting churn.
- Your conventions first. I follow your contributing guide, code style, and commit format, and I sign the DCO or CLA if you ask.
- A description that stands alone. It covers the problem, the change, why it helps users, how I tested it, and what it doesn't cover.
- Proof included. Tests for behavior, plus screenshots or a recording for anything visual.
- An issue first for big changes. That way we agree on the approach before I write the code.
- I stay for review. I address feedback and follow up on anything I ship.
Recently merged upstream: two PRs to Looms, a TypeScript SDK for durable agents. One adds a generic evaluation kind (#5); the other adds defineKind and createKind for custom types (#6).
TypeScript Node.js React Next.js Astro Electron Hono Cloudflare Workers Durable Objects MCP MCP Apps OpenTelemetry Vercel AI SDK OpenAI · Anthropic · Gemini OpenAI Realtime Vitest Playwright GitHub Actions pnpm · Turborepo
San Francisco · Happy to talk about agents, MCP, and developer tools: reachjalil@gmail.com