Agent session logs are stored and searchable. You can even download the raw logs and push your local Claude Code sessions to the platform.
One live view shows every session across your org: what's running now, what shipped, and what failed, with durations and outcomes by agent and repo.
Every session across the org, live
Each session reports its exact cost, broken down by tokens, CPU, and memory. Chart spend over time and see it by agent and by repo, so you always know where the money goes.
Spend by agent, tokens, CPU, and memory
Follow any session in real time: steps stream to the CLI or dashboard as they happen, and a live summary says what the agent is doing right now. The same WebSocket feeds your own tools through the API.
Steps stream to the CLI as they happen
When a session needs a closer look, open it: the agent's thinking, the tools it called, the tests it ran, and the diff it shipped, with per-step token counts. Every raw transcript is kept as a durable record, so Ellipsis is the system of record for your agent work.
The full step timeline, kept permanently
Transcripts, recaps, and PR references are all indexed, with semantic search alongside full text, so did-anyone-look-into-this is a query, not archaeology.
Investigated webhook retry storms after the queue migration
Added idempotency keys to the trade-webhook consumer
Semantic search over every transcript
Everything the platform records and keeps, so agent work has the same paper trail as the rest of your infrastructure.
Assistant output, each tool call with its input and result, and per-step token counts, recorded live.
Total spend broken into model tokens, sandbox CPU, sandbox memory, and the fee, with per-turn timing.
The exact config the session executed. Later edits never change what a past session shows.
The person or event that started every session, so attribution is on the record.
A budget cutoff reads differently from a crashed tool call: budget_hit, tool_call_failed, and friends.
The complete ordered session history, one JSONL file from the CLI or API.
Sessions stream to the dashboard, the CLI with --watch, and your own tools over WebSocket.
Message a running session and it redirects without restarting; stop it from the dashboard or CLI.
Semantic search alongside full text across transcripts, recaps, and PR references.
Transcripts keep indefinitely by default; admins can set retention from 7 to 365 days.
Everything the dashboard shows is readable over the REST API: sessions, steps, costs, and logs.