Kolena

Create a Run from local documents: name every file in `filenames`, then upload them to the `upload_url` this returns. The upload creates the Run and returns it. Send them as a single `multipart/form-data` POST to `upload_url`, with one part per file, every part named `files`, and each part carrying its filename from `filenames` — for example `curl -X POST "$upload_url" -F "[email protected]" -F "[email protected]"`. Parts are matched to `filenames` by name, so order does not matter, but the two sets must be identical. Each part's filename is a bare name, never a path. A Run holds at most 300 documents, none over 250MB. A `.zip` counts as its contents: each file inside becomes its own document under the same limits. The target lasts 10 minutes and is consumed by a successful upload. After a failed upload, fix what the error reports and post again; the error says if you need a new target. `upload_url` is a direct URL you must post to yourself; it is not proxied through this connection, and it carries its own credential, so send no authorization header. If you cannot make outbound requests, say so — the user must create the Run from the Agent's page in Kolena instead. The Run comes back still processing. Poll kolena_get_run until its status settles, then read the Agent's output with kolena_list_run_results.

kolena_create_upload_target

Create an empty Agent in a Kolena Workspace. The new Agent has no Prompts and therefore performs no processing until Prompts are added with kolena_create_prompt. Returns the created Agent summary, including its ID for subsequent Agent and Prompt operations.

kolena_create_agent

Create a Folder in a Kolena Workspace. The Folder may be created at the Workspace root or inside one existing top-level Folder, and existing Agents may be moved into it during creation. Returns the created Folder.

kolena_create_folder

Create a Prompt in a Kolena Agent. A Prompt is one step the Agent runs over each Run's documents; see `config` for the available types. The new Prompt is appended to the Agent's configured Prompt order. References in its instruction establish which Prompts it reads from, and therefore when it runs; Document Filters determine which documents it receives; run conditions determine whether it runs at all. Returns the created Prompt.

kolena_create_prompt

Permanently delete a Prompt from a Kolena Agent, along with every result it has produced across that Agent's Runs. This cannot be undone. If other Prompts reference this one, deletion is refused and the response names them — update those first. To change what a Prompt does without losing its results, use kolena_update_prompt instead. Returns the deleted Prompt's id, name, and type as confirmation.

kolena_delete_prompt

Get a time-limited link to a Run's documents, always as a zip. `filenames` lists its entries, by original filename; documents sharing an original file collapse to one entry, so it may be shorter than that Run's `documents`. `download_url` is a direct URL you must fetch yourself. It is not proxied through this connection and stops working at `expires_at`. If you cannot make outbound requests, do not guess at the documents' contents — tell the user and give them the Run's `web_url` from kolena_get_run.

kolena_download_run_documents

Billed credit spend for the Account: one entry per calendar month, from the Account's first billed usage through the current month (months without usage show zero), each split by funding source — monthly quota, prepaid, and overage beyond both — and broken down by Workspace. Spend covers only the Workspaces whose usage the user can see. Also returns the current month's quota and every prepaid grant, including revoked ones, each with credits granted and used; these are Account-wide. Start here for spend or remaining-credit questions; use kolena_list_usage_records only to drill into the entries behind a figure.

kolena_get_billing_summary

Retrieve one Run. Prompt results are not included — use kolena_list_run_results for those. `status` is `running` while the Agent is still working, and settles to `success` or `failed`.

kolena_get_run

Retrieve one accessible Agent, including the Agent Guidelines and Agent settings, and its ordered Prompts — each with its instruction, type-specific configuration, model and processing settings, Prompt Documents, the Prompts it reads from, the Document Filters scoping its inputs, and the conditions gating whether it runs. Trashed Agents are not returned. Use this after identifying the intended Agent with kolena_list_agents.

kolena_get_agent

List the Kolena Accounts the authenticated user can use here. Call this first: every other tool requires one of these Accounts' id as account_id. If there are several, ask the user which to use.

kolena_list_accounts