Hive
Hive documentation / Capabilities

Pipelines and automation

Chain agents into repeatable workflows and trigger them with schedules or webhooks.

This chapter explains how to chain agents together and trigger them without manual intervention.

1. Pipelines

Pipelines are static, deterministic graphs of agent executions. Unlike Colonies (which use an LLM orchestrator to dynamically plan steps), Pipelines always execute the exact steps you define.

Execution Modes

  • Sequential: Step 2 waits for Step 1 to finish. The raw string output of Step 1 is automatically injected into the prompt of Step 2.
  • Parallel: Step 2 and Step 3 run at the same time. This is useful for fan-out tasks (e.g., asking three different models to review the same codebase simultaneously).

Data Flow

When a pipeline runs, the system tracks the final_output of each step. If Step A feeds into Step B, Step B’s prompt evaluates with the {input} variable replaced by Step A’s output.

2. Schedules (Cron)

Schedules allow you to run an agent automatically on a timer.

  • Storage: Schedules are stored in the SQLite database and loaded into memory on boot by server/lib/scheduler.js.
  • Syntax: Uses standard 5-field cron syntax (e.g., */15 * * * * for every 15 minutes).
  • Natural Language Parsing: The UI provides a natural language to cron parser, but underneath, it always saves standard cron strings.
  • Concurrency Prevention: If a cron fires (e.g., every 1 minute) but the agent’s previous run is still executing, the scheduler will skip the new tick to prevent duplicate cascading runs.

3. Webhooks

Webhooks allow external systems (GitHub, Zapier, custom scripts) to trigger agents or pipelines in Hive.

The Webhook Schema

When configuring a webhook action, you specify a prompt template that can include {input} (the raw JSON payload) and {context} (projected specific variables).

Example Github Issue Payload:

{
  "action": "opened",
  "issue": {
    "title": "Fix the login bug",
    "body": "Users cannot log in when using Safari."
  }
}

Context Projection (webhookProjection.js): Hive attempts to flatten and extract the most relevant data into a context object to save tokens. In the prompt template for your webhook action, you can write: Please review this new issue: {input}

Security & Rate Limiting

  • Signatures: Hive validates incoming webhooks using constant-time string comparisons against the generated secret.
  • Runaway Cost Protection: The environment variable HIVE_MAX_TRIGGERED_COLONY_RUNS (default 2) caps the number of concurrent Colony missions that can be triggered by a webhook flurry, preventing a misconfigured GitHub action from draining your API credits.