› judged by model, keep fit ≥ 75
› bid = $45 − 15% = $38.25/hr
Each agent searches your keywords on Upwork, has your AI model score every job, then submits proposals and answers clients on its own, on your schedule. Want a look first? Turn on review for any tool.
Pick tools from Upwork's official MCP server, set how it bids and how often it runs. From then on it works on its own.
Agents, manual tools and the full Upwork inbox for a flat monthly price. AI usage runs on your own OpenAI key, so you see exactly what the model costs.
Your Upwork inbox, inside Upwork Agents. The agent reads the job, your proposal and the whole thread before it replies, and it can attach files. Read, reply and attach yourself any time.
While I was heads-down on a client build, my agent found an Expo job, wrote the proposal and got it in within the hour. I just read the log afterwards.
How it connects, what it costs to run, and how much control you keep.
Create your first agentThrough Upwork's official MCP server. You sign in with Upwork using OAuth and approve access, so your Upwork password never passes through us. The access tokens are stored encrypted.
Yes. You paste your OpenAI API key once, and your agents use it to score jobs and write proposals and replies. OpenAI bills you directly for that usage, and every run shows how many tokens it used.
Everything: agents with all their tools, bid rules, the Approvals queue, run logs, and the manual tools (Job Posting, Proposals, Conversations and the Dashboard). The only separate cost is your OpenAI usage.
Only if you let it. Each tool can act on its own or wait for your review. With review on, drafts land in Approvals with a fit score and a log of what the agent did, and you can edit them before approving.
With the bid rule you set on each agent: a discount on the client's average rate, plus a fallback rate it won't go below. For example, a client averaging $45/hr with a 15% discount gets a $38.25/hr bid.
On the schedule you pick for each agent, such as every hour, every 2 hours or once a day. On every run, each keyword is searched separately and the AI model judges the jobs it finds.
Yes. Every run is logged with the agent's reasoning, the tools it used, its token usage and the outcome.
Yes. Job Posting lets you search jobs, read full posts and apply with a profile switcher, boost and screening answers. Proposals shows what you've submitted and your drafts, and Conversations lets you read, reply and attach files.
Only you. You sign in with Clerk, and every record is stored under your own user ID.