# hiveloom 1.3.0 documentation

> An agent is a model plus its harness. hiveloom confines an agent to one task by declaring its tools, context, loop policy, budgets, guardrails, and deterministic acceptance checks in a portable folder.

This documentation is designed for direct agent access. Every page below is raw Markdown; fetch https://docs.hiveloom.cloud/v1.3.0/llms-full.txt for a single-context version. The current stable release and available documentation editions are listed in https://docs.hiveloom.cloud/versions.json.

## Start here

- [Introduction](https://docs.hiveloom.cloud/v1.3.0/index.md): Build an agent for a repeatable job. Check its answers. Improve it from evidence.
- [Task confinement](https://docs.hiveloom.cloud/v1.3.0/task-confinement.md): A hiveloom harness declares one job and enforces its operating and acceptance rules.
- [Install](https://docs.hiveloom.cloud/v1.3.0/install.md): hiveloom 1.3.0 requires Python 3.11 or newer and is available on PyPI.
- [Release notes](https://docs.hiveloom.cloud/v1.3.0/release-notes.md): Autonomous improvement, with experiments and a reviewable result. This release adds research programs: a director investigates a harness's failures, while an engine measures candid
- [Quickstart](https://docs.hiveloom.cloud/v1.3.0/quickstart.md): Build a summarizer whose output must be valid JSON with a nonempty summary. Construction, validation, and the dry-run need no model API key. The final run does. Start with the 1.3.
- [Demo harnesses](https://docs.hiveloom.cloud/v1.3.0/demos.md): Use live examples to try a complete task, and scripted lab fixtures to inspect specific runtime mechanisms without model credentials. Each folder contains a spec, supporting assets
- [Evaluations](https://docs.hiveloom.cloud/v1.3.0/evals.md): The checked-in evaluations live in evals/, each with its harnesses, scoring code, and committed results. They address task execution and, in 1.3.0, autonomous improvement. These ar

## Reference

- [Harness spec](https://docs.hiveloom.cloud/v1.3.0/spec.md): The harness spec is the task-confinement contract: a declarative YAML document (harness.yaml) that fixes the task, available capabilities, autonomy limits, acceptance checks, and e
- [Delegation](https://docs.hiveloom.cloud/v1.3.0/delegation.md): A harness is one task, confined. Sooner or later it meets a request that another harness on the same machine does better: a summarizer handed a ranked-retrieval question, a triage 
- [Architecture](https://docs.hiveloom.cloud/v1.3.0/architecture.md): hiveloom starts from a simple definition: agent = model + harness. It confines that agent to one job and makes its success provable. The harness declares the task, tools, context, 
- [Models and providers](https://docs.hiveloom.cloud/v1.3.0/models.md): A harness names the model it executes with in two fields:
- [Extending](https://docs.hiveloom.cloud/v1.3.0/extending.md): hiveloom's catalog is open: tools, guardrails, validators, loop policies, compaction methods, event hooks, eval datasets/scorers, and model providers are catalog entries, and exten
- [Evolution migration](https://docs.hiveloom.cloud/v1.3.0/evolution-migration.md): hiveloom 1.2.0 carries the reusable parts of the ARC-AGI-2 benchmark work into the library. This note records what generalized, what was corrected on the way, and what was delibera

## Build and debug

- [The workbench](https://docs.hiveloom.cloud/v1.3.0/workbench.md): The workbench puts a copilot conversation beside a harness workspace. Ask the copilot to build, run, diagnose, or improve a harness, then inspect the actual spec, run journal, meas
- [Journal and forks](https://docs.hiveloom.cloud/v1.3.0/journal.md): Every run writes an append-only JSONL journal into the harness's .hiveloom/traces/. From 1.0.0 that journal is progressive, self-describing, and tamper-evident — complete enough to

## Measure and improve

- [Local evaluations](https://docs.hiveloom.cloud/v1.3.0/evaluating.md): Hiveloom eval documents keep dataset and scoring policy outside harness.yaml. A scorer runs only after normal harness verification and sees the public RunResult; its numeric output
- [Signal-driven evolution](https://docs.hiveloom.cloud/v1.3.0/signal-driven-evolution.md): Evolution changes a harness because its runs said something. This document covers how hiveloom finds where the runs point, makes every change a prediction about that place, checks 
- [Research programs](https://docs.hiveloom.cloud/v1.3.0/research.md): A research program runs a bounded improvement loop on a harness. You define what better means, which spec fields may change, which tools may execute, and how much the program may s

## Operate

- [Deploying and evolving](https://docs.hiveloom.cloud/v1.3.0/deploying-and-evolving.md): A deployable harness is a folder containing harness.yaml and its declared code/assets. Runs create local .hiveloom evidence; deployment packages exclude that state and credentials.
- [Control plane](https://docs.hiveloom.cloud/v1.3.0/control-plane.md): hiveloom control-plane <harness-dir> exposes a deployed harness's CLI surface over HTTP, bearer-authenticated with the ed25519 keys described below — a selected operational subset,
- [Link/sync protocol](https://docs.hiveloom.cloud/v1.3.0/sync-protocol.md): hiveloom cloud link pairs a local harness directory with a server, and hiveloom cloud sync (or --sync on hiveloom run) keeps the two sides converged. The protocol was built for hiv

## Source

- [Repository](https://github.com/FrancescoMrn/hiveloom)
- [Evaluations](https://github.com/FrancescoMrn/hiveloom/blob/main/evals/README.md)