Pipelines — Scheduled Agents.
A scriptorium of agents working in chorus. Schedule runs, branch on conditions, retry failures, observe every step — all against your encrypted vault.
Agents Working in Chorus
A pipeline is a scheduled agent run with a job to do. Morning digest: summarize every memory written overnight. Weekly research sweep: query the vault for the week's open questions, run a research agent, write findings back. Nightly code review: pull PRs from git, match against coded-in constraints in the vault, comment on what disagrees. Pipelines turn your vault from a passive store into an active collaborator that notices, reminds, and reasons on its own schedule.
Pipelines that work while you sleep.
Schedule an agent against your encrypted vault. Every run becomes a paragraph in the codex — searchable forever.
Schedule it, forget it
Cron expressions or event hooks. Every pipeline shows its next run and a full activity log of past runs.
Branch, retry, fan out
Multi-step with conditions, exponential backoff, and parallel fan-out. Durable across server restarts.
Every run, a paragraph
Results write back as tagged memories. Search 'what did the weekly-research pipeline find?' a month later.
Schedule It, Trigger It, Forget It
Cron expressions for time-based runs. Event hooks for reactive runs (new memory of type X, new note in folder Y, member mention). Manual 'Run now' for testing. Every pipeline shows its next scheduled run + an activity log of past runs. Schedule is a first-class field; change a cron and the next run picks it up automatically. Pause any pipeline with one click — the schedule halts but the definition survives for later.
Branch, Retry, Fan Out
Pipelines are multi-step. Step one runs a search; step two conditions on the result ('if more than 5 new memories, summarize them'); step three writes a note; step four notifies the team. Steps retry on failure with exponential backoff. A step can fan out — run the same agent on N inputs in parallel, merge results. Steps can emit memories that later steps read. The whole pipeline is durable: if the server restarts mid-run, state is restored and execution resumes.
Observability Built In
Every pipeline run produces an audit entry — which steps ran, how long each took, what tokens they cost, what they read, what they wrote. Drill into a run: see every tool call, every retry, every output. Filter by status (succeeded, failed, running). See the cost of each pipeline over the last 30 days. When something breaks, you don't guess — you open the run and read what happened. When it works, you see exactly how it worked. No black boxes.
Templates & The Editor
Start from a template: Daily Digest, Weekly Research, Session Review, PR Triage, Consolidation Pass. Or build from scratch in the pipeline editor — drag steps, configure agents, wire inputs to outputs, test on a sample input before saving. The editor shows you the live graph of your pipeline and validates wiring as you go. When you save, the pipeline is immediately registered with the scheduler. Agents don't need to know anything about pipelines — they just get invoked with the right context and write back to the vault as usual.
Every Run, a Paragraph in the Codex
When a pipeline finishes, its output is not thrown away. Results are written back into the vault as memories, tagged with the pipeline name and run ID. A month later, you can search 'what did the weekly-research pipeline find about our onboarding flow?' and the agent recalls the whole history. The codex grows with the work, because the work IS the codex.
A codex worth keeping.
Free to start. Encrypted always. Connect your first agent in under a minute.