Give the agent your application logs and it can move from 'this code path runs' to 'here is the line where it failed, and what else was happening at that moment.' Logs connect to Seq, are labelled per environment, and are offered to the agent only when a source is configured.
A connected log source closes the loop on an investigation. The agent already reads your code and queries your database; logs let it correlate a failing request with the exact log line, the stack trace, and the surrounding activity — without anyone hand-copying entries into the chat. Codepanion connects to Seq.
Log sources are configured in the app, under Settings → Log sources — you don't push logs to us over an ingest call. Give us your Seq server's base URL and an API key; we test the connection, then store the key as its own encrypted Key Vault secret. The key is never shown again and never returned by the API.
POST /settings/log-sources/{environment}
{
"provider": "seq",
"baseUrl": "https://logs.acme.io",
"apiKey": "<your Seq API key>"
}Most teams never touch the API directly — the Settings page drives this for you, including a Test button that live-probes the source before you save. Managing log sources is owner-only. An optional Seq signal id scopes every query to a saved Seq view. To remove a source, delete it from the same page.
Each source is labelled with an environment — qa, staging, prod — exactly like your database connections and code snapshots. When an investigation is pinned to an environment, the log tools read that environment's source automatically, so the agent stays on one environment end to end: the QA build's code, the QA database, and the QA logs. A tenant with a single source can leave it under the default label and never think about it.
Once a source is connected, two read-only tools appear in the agent's toolbelt. They're offered only when a log source is configured — no dead buttons, no empty searches:
search_logsSearch the environment's logs by free text, minimum level, and time range — "every error mentioning order 88123 in the last hour" — newest first.get_log_contextPull the entries surrounding a moment in time — anchored on a log entry or a raw instant, widened by a window like 5m, optionally narrowed to a single distributed trace. The "what else was happening then" view.Both surface as plain-language tool steps in the investigation, and every call is recorded in the audit log against the environment it ran in. The agent never sees the API key — it's resolved from Key Vault per call.
Good to know
Connecting logs is what turns a "why is checkout failing in QA?" question into a complete answer — the failing code path, the offending row, and the log line that ties them together. It's how QA and engineering use the agent the same way support does.
If your Seq server lives inside your network and must stay there, the on-prem connector reaches it for you. Mark the environment's log source as connector-mode and the agent routes search_logs and get_log_context over the connector's outbound channel — the same one it uses for the database. Codepanion holds no credential: the connector reaches your log store using a node-local key or the host's ambient cloud identity, on your side of the channel.
Every pilot customer gets hands-on onboarding from the founding team. We'll walk through setup together and make sure everything is working.