Skip to main content
OpenWork is an open source desktop app that runs an AI agent on a folder on your computer. You point it at a workspace, describe a task, and the agent reads and writes the files in that folder. The source lives at github.com/different-ai/openwork. OpenWork ships with a built-in Requesty provider. Connect it and you get:
  • 600+ models in the OpenWork model picker: Claude, GPT, Gemini, Mistral, DeepSeek and more, all with one API key.
  • Per-task visibility: every request the agent makes shows up in Logs and Analytics with its model, provider, tokens, cost and latency.
  • Spend limits on the key OpenWork uses, so a long-running agent can never go past a budget.
  • Routing policies such as Fallback, and EU routing for teams that need data to stay in the EU.
OpenWork runs its agent on OpenCode, so the same opencode.json configuration format applies. This guide was tested with OpenWork v0.18.67.

Prerequisites

  • A Requesty account and an API key from the API Keys page. We recommend a dedicated key for OpenWork, for example openwork, so its usage and limits are separate from your other tools.
  • A folder you want OpenWork to work in.

Install OpenWork

1

Download the app

Go to openworklabs.com/download and pick the build for your system: macOS (Apple Silicon or Intel), Windows (x64 or ARM64) or Linux (AppImage or tar.gz). Every build is also attached to the GitHub releases.
OpenWork download page with macOS, Windows and Linux builds
On Linux, make the AppImage executable and start it:
2

Add your folder as a workspace

OpenWork opens on a default OpenWork Chat workspace. Click Add workspace, choose Local workspace, pick your folder in the Authorize folder dialog and click Create Workspace.
OpenWork Local workspace dialog with a folder selected and the Create Workspace button

Connect Requesty

1

Open AI Providers

Open the account menu at the bottom left, go to Settings > AI Providers and click Connect a provider. Search for Requesty and click Connect.
OpenWork Connect a provider dialog with Requesty in the search results
2

Paste your API key

Paste your Requesty API key and click Save key. Keys added here stay on this device.
Requesty API key dialog in OpenWork with a masked key and the Save key button
3

Check the connection

Requesty now appears under On this device in AI Providers, with a Disconnect button. Click Back to app to return to your workspace.
OpenWork AI Providers page listing Requesty as connected on this device
4

Pick a model

Start a New session, click the model name under the task box and search the list. Requesty models are grouped under Requesty. Pick one, for example Claude Sonnet 5.5.
OpenWork model picker filtered to sonnet 5.5, listing Claude Sonnet 5.5 and other models from Requesty
These models are Requesty Managed Policies such as claude-sonnet-5-5 and gemini-3.8-flash, so Requesty chooses the provider for you. You can switch models per session.

Run your first task

Describe a task that uses the files in your workspace, then click Run task. For example, in a folder of invoice PDFs:
The agent reads the files, writes the result back into the folder, and links the new file under its reply. Click the file to open it in the preview panel.
OpenWork reply with an invoice table, a grand total and a flagged duplicate invoice

See every request in Requesty

A single OpenWork task is usually several model requests, as the agent reads files, calls tools and writes output. All of them appear in Requesty.
1

Filter Logs to your OpenWork key

Open Logs and click your OpenWork key in the API key column, or use Add filter > API key. Each row shows the provider, model, routing type and policy. Use Columns to add tokens, cost and latency.
Requesty Logs filtered to the OpenWork API key, showing Claude Sonnet 5.5, Gemini 3.8 Flash and GPT-6 Luna requests
2

Open a request

Click a row to see first-token latency, input, cached and output tokens, the routing decision and the provider that served it.
Requesty request detail for claude-sonnet-5-5 served by Bedrock, with latency, token and routing breakdown
Agent turns resend much of the same workspace and tool context, so a large share of input is often served from cache, as in this request.
3

Compare models in Analytics

In Analytics, filter by your OpenWork key and set Group to Model to see requests, tokens and cost per model.
Requesty Analytics model usage table for the OpenWork key with requests, tokens and cost per model

Set a spend limit

An agent that works through a large folder can make many requests. Put a monthly limit on the key OpenWork uses so it can never go past your budget.
1

Open the key's limit

On the API Keys page, click the value in the Usage column of your OpenWork key (it reads No limit by default) to open Update API Key Limit.
2

Enter the limit

Enter the Monthly Limit (USD) and click Update Limit. The smallest limit is $1.00.
Update API Key Limit dialog with a monthly limit of $20
The Usage column then shows spend against the limit for each key:
API Keys list showing one key at $1.08 of a $1.00 limit and another at $0.64 of a $20.00 limit
When a key reaches its limit, Requesty blocks new requests and OpenWork shows the error inline as the agent’s reply:
OpenWork session where the reply is the Requesty monthly spend limit message
Requests that are already running when the limit is reached still finish, so usage can end slightly above the limit, as in the $1.08 / $1.00 example above. To continue, raise the limit on the API Keys page.
A routing policy gives OpenWork one stable model name while you decide on the Requesty side which models sit behind it. Combined with the EU endpoint, you can keep every request in the EU and still fail over to a second model.
1

Create the policy

Open Routing Policies, go to the Custom tab and click Create policy. Give it a name, for example openwork-reliable, choose the Fallback strategy and use Add model to add models in the order they should be tried. To stay in the EU, add EU-pinned models such as bedrock/claude-sonnet-5-5@eu-west-1 first and azure/gpt-6.1-sol@swedencentral as the fallback.
Requesty routing policy openwork-reliable with the Fallback strategy, Claude Sonnet 5.5 in eu-west-1 first and GPT-6.1 Sol in Sweden Central second
After you click Create policy, its model ID is shown under the name: policy/openwork-reliable.
2

Store the API key in a file

Save the Requesty key you want this provider to use in a file that only you can read:
In OpenWork v0.18.67, variables saved under Settings > Environment were not passed to the agent, so {env:...} references to them failed. Use a {file:...} reference as shown here.
3

Add the provider to your workspace

Create opencode.jsonc (or opencode.json) in the root of your workspace folder:
opencode.jsonc in a text editor with the Requesty EU base URL, a file-based API key and the policy model
Every policy has to be listed under models. OpenWork only offers the models it knows about, so a policy that is not declared here does not appear in the picker.
4

Pick the policy and run a task

OpenWork picks the file up without a restart. Start a new session, search the model picker for the name you gave the policy, and select it under Requesty EU.
OpenWork model picker showing Reliable (Claude, GPT fallback) under Requesty EU
Run a task as usual:
OpenWork answer produced by the Reliable (Claude, GPT fallback) policy
5

Confirm the region in Logs

In Logs, open one of the requests. The detail shows Policy: fallback, the policy name, the model that served the request and its region, here claude-sonnet-5-5@eu-west-1 on Bedrock.
Requesty request detail showing Policy: fallback, openwork-reliable, claude-sonnet-5-5@eu-west-1 and Bedrock
To change models later, edit the policy in Requesty. OpenWork keeps using policy/openwork-reliable and the next request follows the new configuration. If you only need EU routing without a policy, list EU models such as claude-sonnet-5-5@eu under models instead. To use a policy on the global endpoint, set baseURL to https://router.requesty.ai/v1.

Troubleshooting

Open Settings > AI Providers, disconnect Requesty and connect it again with a valid key. For a custom provider in opencode.jsonc, check that the key file exists and contains only the key.
Declare it under models in opencode.jsonc in the workspace root, using the exact ID from Routing Policies or the models list. Also check that the model is enabled in your organization’s Approved Models.
The model or policy ID is not known to Requesty. Check the spelling and the policy/ prefix against Routing Policies or the models list.
The key OpenWork uses is over its limit. Raise or remove the limit on the API Keys page. See Set a spend limit.

References

Last modified on November 19, 2026