Automation

Work While You Sleep.
Literally.

Schedule agents to run overnight. Set webhook triggers. Define event-driven workflows. Your agents keep working long after you clock out.

Scheduling & Cron

Schedule anything on any schedule

One-time messages, recurring cron jobs, or autonomous background loops. Your agents can claim pending tasks, deliver scheduled messages, and run completions on a 60-second tick — entirely without you.

Cron Scheduling

Full 5-field cron expressions. Every morning at 6 AM, every Monday, or every 5 minutes.

One-Shot Messages

Schedule a prompt to fire at a specific datetime. Set it and forget it.

Autonomous Loop

Background agents poll for tasks every 60s. Self-running AI workers.

Policy Escalation

Auto-escalate blocked tasks. Auto-checkpoint. Deadline marking.

[Screenshot: schedule modal showing cron config, one-shot datetime picker, and scheduled message list]
Chain PipelineInputRaw dataStage 1ClassifierIntent detectionStage 2ExtractorEntity parsingStage 3GeneratorOutput craftedOutputAggregation: Structured result with chain provenance! prompt template → auto-filled per stageError → re-route to fallbackEach stage is a reusable building blockChain stages can be agents, tools, workflows, or sub-chainsOutput of stage N → input of stage N+1 with auto-prompt generation
Workflows

Codify your AI process into repeatable workflows

Define declarative, versioned workflows with steps, input validation, and artifact tracking. Run them with --detach and check results later. Replay them with different inputs.

Declarative Specs

Define workflows as versioned TOML or JSON. Git-track your AI processes.

Step Execution

Each step is an agent call with validated inputs and outputs.

Replayable

Run the same workflow with new inputs. Compare results across runs.

Run History

Every run is tracked. Status, artifacts, and timing — all auditable.

Hooks & Triggers

Guardrails and event-driven extensions

Hooks intercept execution at key points — pre-tool-call, pre-merge, pre-message. Use them to validate, block, transform, or enrich. Attach bindings to fire agents on specific events.

Risk Policies

Block SideEffect tools in NonInteractive mode. Simulate dry runs for validation.

Webhook Triggers

POST from GitHub, Stripe, or n8n to fire a prompt.

Binding System

When THIS event fires, at THIS hook point, run THIS agent. Persisted contracts.

External Actors

Dispatch to external HTTP endpoints. n8n, Zapier, or custom services.

Hook System ArchitectureTrigger Pointson_messageon_responseon_branchon_errorBinding LayerHook → Agent routingFilter conditionsPriority orderingTimeout configObserving AgentsMeta Agent 1Log monitorMeta Agent 2Quality checkMeta Agent 3Re-route fallbackDispatchedActions→ Intercept message→ Modify context→ Terminate branch→ Re-route agent→ Trigger hook→ Log auditHooks enable self-observing AI systemsAny trigger point can bind to any meta agent. Meta agents observe, intercept, or modify the primary flow.Use cases: moderation, quality gating, cost control, audit logging, automatic re-routing on failure.
Variable Fan-OutPrompt Template"Write a personalized emailfor {firstName} about{topic}. Mention theirrole {role}..."CSV / Data SourcefirstName,topic,roleAlice,AI,EngineerBob,ML,Scientist ...Fan-OutEngine500 iterationsEmail 1: Alice → AIPersonalized outputEmail 2: Bob → MLPersonalized outputEmail 3: ......Aggregated Result500/500 emails generatedOne template, hundreds of personalized outputsPrompt template variables auto-filled from CSV rows. Each row → one unique generation.Batch processing with configurable concurrency, error handling, and progress tracking per row.
Variable Fan-Out

One prompt. 500 personalized outputs.

Define variables in your prompts with Handlebars templates. Import values from CSV files. Bonsai fans out a single prompt template across every row — automatically.

CSV Import

Upload a CSV. Select columns by name or index. Variables populate automatically.

Template Variables

Handlebars syntax. Full logic support — loops, conditionals, helpers.

Bulk Execution

One command fans out across every variable row. Track all outputs.

Preview Mode

Preview CSV headers and column values before you fan out. No surprises.

Wake up to completed research.

Schedule your research agents at 11 PM. Open your laptop at 7 AM to a synthesized report, not a blank cursor.