> ## Documentation Index > Fetch the complete documentation index at: https://felan.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Sessions > Start Felan work, inspect live progress, and continue conversations. A session is the dashboard record for a Felan request or automation run. It keeps the prompt, transcript, status, and delegated agent activity together. When a request needs shell, filesystem, repository, or browser access, Felan routes those tools into a separate agent workspace. See [Security architecture](/docs/security) for the runtime and workspace boundaries. ## Start a session Select **New Session**, enter a request under **What should Felan do?**, and submit it. Felan creates the session, shows startup progress, and opens its detail view. While the first response runs, Felan asynchronously creates a concise title from the initial request. Until the title is ready, or if it cannot be created, the session uses a short preview of the request as its display title. A title you rename remains authoritative, and later follow-ups do not replace it. Scheduled and event-driven sessions retain their automation or event title. Write requests as outcomes with useful constraints. For example: ```text theme={null} Review the open change for authentication regressions, run the relevant checks, and report findings by severity. Do not change files. ``` ### Start a session through the API Use a team API key to create the same kind of root conversation from an external tool: ```bash theme={null} curl -X POST "https://app.felan.ai/api/v1/teams/{team-slug}/sessions" \ -H "Authorization: Bearer bzy_team_abc123..." \ -H "Content-Type: application/json" \ -d '{"prompt":"Review the open change and report authentication regressions."}' ``` The API returns `202 Accepted` with a public session summary. The initial status is `pending`; poll the session detail until it becomes `running` or a terminal status. Session summaries contain: * `id` and lifecycle timestamps * `kind`, `parent_session_id`, and `persona_id` * `prompt`, generated `title`, and `status` * `llm_cost_usd`, `num_turns`, `num_tool_calls`, and `model` Session detail adds `children`, containing summaries for direct delegated children. Runtime metadata, credentials, internal session keys, container identifiers, log paths, and private errors are not part of the public DTO. ## Follow live work The session detail view shows: * A title that can be renamed * The person who submitted the initial request * The current status and elapsed duration when available * The live transcript * Errors when a run fails * Delegated subagent activity associated with the root session * A **Kill** action while the selected session is running Session statuses include `pending`, `running`, `completed`, `failed`, `timed_out`, and `cancelled`. Pending and running sessions can produce more work. Completed, failed, timed out, and cancelled sessions are terminal. ## Poll session progress through the API Use `GET /api/v1/teams/{team-slug}/sessions/{session-id}` for current status and direct child summaries. List sessions with `GET /api/v1/teams/{team-slug}/sessions`; the optional `status` filter accepts a session status or `all`, and `limit` ranges from 1 to 100. Read transcript events with: ```text theme={null} GET /api/v1/teams/{team-slug}/sessions/{session-id}/transcript?after_sequence={cursor}&limit=100 ``` Events are ordered by `sequence`. Both event sequences and `next_cursor` are decimal strings, preserving values larger than JavaScript's safe integer range. Start without `after_sequence`, process the returned events, and pass `next_cursor` into the next request. Continue immediately while `has_more` is `true`; otherwise poll again while the session is pending or running. The transcript limit ranges from 1 to 1,000 and defaults to 100. Each event includes a sanitized `raw_event` projection for message and tool content. Provider, credential, request, signature, sender identifier, and repeated conversation metadata remain internal. ## Continue the conversation Running or completed root conversational sessions provide a follow-up composer below the transcript. Use it when the next request depends on the same conversation and workspace context. Start a new session when the work is unrelated or should have an independent history. The public API provides two different ways to influence work: * `POST /messages` adds a conversational follow-up to a running or completed main root session. It preserves the conversation and workspace context. * `POST /steer` changes the direction of a currently running target without creating a new conversational turn. Both bodies use `{ "message": "..." }`. Use `POST /kill` to request cancellation of a running target. Accepted actions return `{ "data": { "accepted": true, "session_id": "..." } }`. ## Repository and network context API-created sessions inherit the team's dashboard configuration. Felan uses the team's connected source-control providers and authorized cloud repositories when repository work is requested. Environment settings and Private Connectivity are also inherited; configure them for the team in the dashboard before starting the session. The session create request itself contains only the prompt. ## Session retention Session prompts, transcripts, delegated activity, and stored session artifacts follow the team's current plan retention window: 30 days for Individual and 90 days for Trial and Team. The window is measured from each session tree's last activity. The complete terminal tree must be strictly older than the cutoff. Felan checks the current plan before atomically deleting the root, delegated sessions, and transcripts. Exact session-scoped Agent Storage artifacts are deleted after the database transaction on a best-effort basis. Conversation summaries, financial usage records, and unrelated team data follow separate policies. Completed database deletions are irreversible. Scheduled retention enforcement deletes eligible sessions automatically; operators can run a non-mutating report first and must monitor storage-cleanup failures. Production operation requires verified 30-day retention for the shared Axiom logging dataset. ## Find previous sessions The sidebar shows recent sessions. Select **All sessions** to open the full list, where you can: * Search by session title, prompt content, agent persona, or status * See status and last activity * Keep each session's position stable while several sessions produce live activity * Distinguish conversational, delegated, scheduled, and event-originated work by its origin indicator * Open a session to inspect the transcript Turn repeatable instructions into a scheduled or event-driven session.