Skip to main content
Use chat to send messages to a Strix agent and get replies. You can attach repositories and files to give context. Chat is available on the Cloud plan and on Enterprise plans.

​
What the chat agent can do in Strix

The agent can run the operations of the Strix API for you: list, read, create, update, and delete scans, findings, schedules, assets, domains, repositories, integrations, knowledge entries, members, webhooks, and pull request reviews. It can also read the agent traces and repositories of your workspace to answer questions about them, and read the Strix documentation to answer questions about the product. The operations are grouped into tools per API area (for example strix_scans, strix_test_users). The agent calls an area tool with an operation name and an arguments object, and asks strix_api_operation_schema for the exact argument shape of an operation when it needs it. The agent acts as the member whose message it is working on. Your role and your asset scopes apply to every operation, so a member with the viewer role cannot make changes. When another member sends a message, the current turn stops and the agent continues as that member. A subagent acts only when a member messages that subagent directly; it then acts as that member. If you use an API token to start a chat or send it a message, the operation scopes of that token also apply: when you message through a token without vulnerabilities:write, the agent cannot change findings. The agent cannot manage chats, API tokens, billing, or credentials. Strix records each change the same as a change in the dashboard.

​
Approval for dangerous operations

Some operations wait for a member to approve them before they run: deletes, cancel, revoke, disconnect, remove, share, changes to member roles or the organization, and actions that start work or reach outside Strix such as scans, retests, fix pull requests, ticket pushes, and member invites. When the agent calls one of these, the chat shows an Approve / Deny card. Nothing runs until a member clicks a decision. An approval expires 30 minutes after the agent requests it. API users read and decide approvals through the API:
Required scope: chat:read
Required scope: chat:write The body decision accepts approve or deny. On approve the operation runs as the token’s user inside the token’s scopes, and the result arrives in the chat as a new message. A request on an approval that is decided or expired returns HTTP 409.

​
Credits

On the Cloud plan, Chat uses prepaid credits from the same wallet as scans. One credit costs $1.00. Chat bills 1.2 credits for each dollar of model cost. Strix deducts the credits after each agent tool call, not after each message. Enterprise chats do not consume credits. Read the spendable Chat balance with GET /api/v1/billing/credits. The chat_balance field gives whole credits only. A credit that Chat has only partly used counts as one.

​
Chat credit usage

A chat uses part of a credit for each agent step. Read the exact usage with GET /api/v1/chat/usage.
Required scope: chat:read The response gives amounts in microcredits. One credit is 1,000,000 microcredits.
  • available_microcredits: the credits that Chat can still spend.
  • typical_chat_microcredits: the median total cost of the recent chats in the workspace, or null when the workspace has no charged chat yet.
  • sampled_chats: the number of chats in that sample.
  • chat_microcredits: the total that the chat in chat_id has used. Zero when you omit chat_id.
  • chat_auto_topup_credits: the whole credits that automatic top-up bought for that chat.
  • chat_auto_topup_cap_lifted: true when an admin lifted the per-chat automatic top-up limit for that chat.
When the workspace has no credits, POST /api/v1/chat and POST /api/v1/chat/<CHAT_ID>/message return HTTP 402 Payment Required with code chat_credit_limit_reached. The request starts no turn and consumes no credits. The body has the same hint, topup_url, and topup_options fields as the scan paywall response. When automatic top-up is on and did not refill the wallet, the body also has a reason field. See Billing and credits for the recovery steps. When the wallet runs out during an agent turn, Strix pauses the chat and writes a system notice to the thread. Buy credits, then send a message to continue. If automatic top-up is enabled for the workspace, Strix refills the wallet before it pauses the chat.

​
Lift the per-chat automatic top-up limit

When the reason is chat_cap_reached, the automatic top-ups for this chat reached the chat_cap_credits limit of the workspace. An admin can lift that limit for this one chat. The chat can then buy credits automatically past the per-chat limit. The monthly cap of the workspace still applies. In the dashboard, the out-of-credits dialog shows the Lift the limit for this chat action. Strix sends the held message when the limit is lifted.
Required scope: billing:write The response contains chat_id and lifted. Send {"lifted": false} to apply the workspace limit to the chat again. The request requires an admin role on the workspace. A chat that the token cannot reach returns HTTP 404.

​
List chats

Required scope: chat:read

​
Start a chat

To attach files, send multipart/form-data instead of JSON. Use the same fields and add the file parts. Required scope: chat:write

​
Get a chat

Required scope: chat:read

​
Send a message

To stop an active agent turn before you send the message, set "cancel_current": true. message can be omitted for a control-only request when either cancel_current or stop_agent is true. cancel_current cancels the in-flight turn; stop_agent parks the target agent and its descendants:
Required scope: chat:write

​
Export workspace files

Chat artifacts live under /workspace in the running agent sandbox. List exportable files:
The listing is capped at 10,000 entries and 2 MiB before the sandbox response is parsed and normalized. An oversized listing returns 413. Download one file. The API accepts either an absolute path under /workspace or a convenient relative path such as report.md; relative paths resolve from /workspace. Traversal and absolute paths outside /workspace are rejected. An individual export is limited to 50 MiB.
Or stream the whole workspace as a ZIP:
Archive export preflights the same 2 MiB / 10,000-entry listing limit, a 50 MiB per-file limit, and 512 MiB of uncompressed content, with a 30-second archive deadline. ZIP entries must be safe relative POSIX paths. The service bounds sandbox JSON and base64 responses before parsing or decoding them, streams with response backpressure, and cancels in-flight sandbox work if the client disconnects. An empty workspace returns 404, and oversized listings or files return 413. Invalid listing, export, or file data from the sandbox returns 502; an unavailable, unready, or timed-out sandbox returns 503. A file read failure or actual-byte overflow fails the stream instead of returning a valid-looking partial ZIP. Required scope: chat:read The CLI adds safe file handling and refuses to replace an existing output unless --force is explicit:
Required scope: chat:write