Bello

Simple setup, clear configuration, fully autonomous execution, and safety by design.
Assign the task and walk away. Bello keeps the coder inside a disposable sandbox while an independent, fresh-context supervisor reviews risky actions, catches drift, and manages recovery.
Across three public ProgramBench tasks and three model-effort settings, Bello outperformed Raw Codex in all 9 matched comparisons, increasing average completion from 44.87% to 61.21%. It is ready to take on your most demanding tasks.

---

Motivation

Modern language models can write code, analyze documents, and solve complex problems, yet the model itself remains a generator of the next fragment of reasoning. When assigned a long, multi-stage task, it must simultaneously remember requirements, plan actions, execute them, assess its own progress, notice errors, and decide when the result can be considered complete.

Combining all these functions is unreliable. As the context grows, the model degrades very quickly and begins to hallucinate [1] [2] [3] [4]. Compressing the history partly addresses the context-size problem, but it can lose a critical rule, decision, or prohibition. Meanwhile, a confident model response is not evidence that the task has actually been completed.

Bello moves the management of complex work to a level above the language model.

In our architecture, a single model is not expected to represent the entire thinking process. We treat a language model as a powerful but limited executor of cognitive operations. Planning the overall process, assigning roles, managing memory, evaluating effectiveness, and making the final decision about readiness should belong to a separate system. Bello implements such a system: not a longer chain of reasoning from a single model, but a reproducible reasoning loop in which a solution is created, reviewed, attacked, corrected, and accepted only after independent confirmation.

Cognitive foundations

Bello's architecture draws on several foundational areas of cognitive psychology.

In Allen Newell and Herbert Simon's heuristic search model, problem solving is viewed as moving from the current state to a goal state through a sequence of operations and subproblems. A person compares the current state with the desired one, chooses an action that reduces the difference, evaluates the result, and restructures the search if the chosen strategy does not work.

Barry Zimmerman's research on self-regulated learning describes activity as a recurring cycle of forethought, performance, and self-reflection. The outcome of an evaluation does not end the process; it changes the plan for the next attempt.

Research on metacognition by John Flavell, Thomas Nelson, Louis Narens, and other authors distinguishes between performing a task and managing that performance. One level solves the task; another observes the work in progress, evaluates confidence and evidence, and decides whether to continue, verify, change strategy, or stop.

Similar patterns have also been found in research on writing. Linda Flower and John Hayes's model describes writing not as a linear sequence of “plan — draft — edit,” but as a recursive interaction between planning, translating ideas into text, and reviewing. A problem discovered while reading may require more than a local edit: it may call for returning to the goal, structure, or original intent.

The common principle behind this work is that producing a result, monitoring the process, and evaluating it critically should not collapse into a single indistinguishable operation. Reliable reasoning requires cycles, specialized functions, and the ability to revisit earlier decisions.

Bello turns this structure into an executable system.

How Bello solves tasks

The process begins by building the first complete solution. The developer agent analyzes the task, modifies the project, runs checks, and creates a working prototype.

The result is then passed to completion review. This component does not continue development or take the author's report at face value. It independently reconstructs the task's mandatory requirements and checks:

• whether the required behavior has been implemented;
• whether the checks support the claimed result;
• whether any modes or edge cases remain untested;
• whether any regressions have been introduced;
• whether fresh validation was performed after the latest substantial changes.

If a problem is found, the work returns to the developer. After the fix, a new full review is performed because a local change may affect other parts of the system.

Once the solution has passed several development and review cycles, the adversary is launched. Its job is not to confirm the work, but to try to break it. It explores invalid inputs, unexpected action sequences, interactions between features, boundary states, and assumptions that the developer and reviewer may have overlooked.

The adversary works independently of the solution's development history. It evaluates the final artifact, not how convincing the author's explanation is. If it finds a potential defect, the solution is sent back to completion review, which determines whether the observed behavior is a genuine violation of the requirements.

Bello therefore implements the following cycle:

build a solution → independently review completeness → fix defects → perform adversarial testing → reassess → accept the result.

Why this structure is a natural fit for software development

Software development demonstrates the limitations of single-pass reasoning particularly well.

A programmer initially builds an implementation around the core functionality. Even a strong first version may fail to account for rare inputs, error recovery, compatibility, operation order, or interactions between multiple components. It is also difficult for authors to evaluate their own code independently: they know what they intended to implement and therefore tend to mentally fill in what the program does not actually contain.

Completion review corresponds to rigorous code review and acceptance auditing. It evaluates the implementation's compliance with the requirements, not the elegance of its explanation. Passing a handful of visible tests is not considered sufficient if they do not cover the full required behavior.

The adversary corresponds to fuzzing, property-based testing, penetration testing, red teaming, and the work of an independent quality engineer. It does not seek confirmation of the standard scenario; it looks for conditions under which the system violates its contract.

Bello tracks not only the quality of the final code, but also the effectiveness of the development process. If the agent repeats the same mistakes, ignores feedback, loses sight of the original goal, or becomes stuck in a flawed interpretation, the system can stop the current line of work, preserve confirmed facts, and start a new pass with a clean context.

This separates the accumulated knowledge about the task from an individual agent's unsuccessful reasoning trajectory.

The same structure in other forms of intellectual work

Similar cycles are used far beyond programming: in mathematics, scientific research, engineering, law, and strategic planning.

In all these fields, a reliable result emerges not from one long sequence of thoughts, but from the interaction between a creator, a reviewer, and a skeptic.

Relationship to existing LLM research

Individual parts of this approach have already been tested in language-model research.

Self-Refine showed that a cycle of generation, critique, and refinement can substantially outperform a single-pass response. Reflexion demonstrated the value of retaining lessons from previous attempts. CRITIC connected self-correction with external tools and observable evidence. Research on self-debugging confirmed that models can improve code by analyzing execution results. Work on adversarial testing and verifier-guided search showed that a separate verifier or attacker can detect seemingly correct solutions that pass conventional checks.

These findings support individual elements of Bello's architecture. Most of this work, however, studies a single mechanism: reflection, correction, verification, debate, test generation, or adversarial search.

Bello combines these mechanisms into a unified system for managing long-running work.

Results

Key findings

• Across all three tasks and model–effort settings, Bello achieved the higher completion score in 9 of 9 matched configurations. The overall unweighted mean increased from 44.87% to 61.21%: +16.33 percentage points (+36.40% relative).
• With GPT-5.6 Sol, Bello achieved the higher completion score in 6 of 6 matched configurations. The unweighted mean increased from 48.92% to 67.04%: +18.13 percentage points (+37.06% relative).
• In the complete GPT-5.6 Sol ultra comparison, every task improved by 18.17–24.59 points, and the macro average increased from 53.53% to 74.03%.
• With GPT-5.5 xhigh, Bello scored higher on all three tasks; the macro average increased from 36.79% to 49.53%: +12.74 percentage points (+34.64% relative).

Evaluation protocol

We evaluated Bello on three ProgramBench tasks: Solar, Samtools, and Rumdl. Raw Codex and Bello were observed on every task with GPT-5.6 Sol in both ultra and xhigh modes and with GPT-5.5 in xhigh mode. We report the completion score recorded in the completion_pct field and time from the runtime field of the run-level data. Completion scores are rounded to the nearest hundredth of a percentage point. Runtime was not held constant, so the comparison is not compute matched. The final solution patches for all nine reported Bello runs, together with SHA-256 checksums, are available in the public evaluation artifacts folder.

GPT-5.6 Sol

ultra

Task	Raw Codex completion	Bello completion	Difference (pp)	Relative change	Raw Codex time	Bello time
Solar	53.13%	71.30%	+18.17	+34.20%	00:32:33	07:39:17
Samtools	51.86%	70.60%	+18.74	+36.14%	00:36:17	19:25:22
Rumdl	55.60%	80.19%	+24.59	+44.23%	01:40:05	07:44:12
Macro mean / total time	53.53%	74.03%	+20.50	+38.30%	02:48:55	34:48:51

Bold completion values indicate the higher observed score within each matched row.

Across the three matched ultra runs, Bello increased completion by 18.17–24.59 percentage points on every task. The unweighted macro average rose from 53.53% to 74.03%, a gain of 20.50 points (38.30% relative).

Figure 1a. Bello-minus-Raw completion differences for the three GPT-5.6 Sol ultra configurations. Every point lies to the right of zero; the diamond shows the unweighted mean difference (+20.50 points). Uncertainty intervals are not shown because each configuration has one observation.

xhigh

Task	Raw Codex completion	Bello completion	Difference (pp)	Relative change	Raw Codex time	Bello time
Solar	46.61%	66.50%	+19.89	+42.67%	00:16:58	04:26:04
Samtools	38.11%	51.93%	+13.82	+36.26%	00:28:48	05:39:13
Rumdl	48.19%	61.74%	+13.55	+28.12%	00:31:57	03:53:35
Macro mean / total time	44.30%	60.06%	+15.75	+35.56%	01:17:43	13:58:52

Bold completion values indicate the higher observed score within each matched row.

All three xhigh tasks improved. The gains ranged from 13.55 to 19.89 percentage points, and the unweighted macro average increased from 44.30% to 60.06% (+15.75 points, +35.56% relative).

Figure 1b. Bello-minus-Raw completion differences for the three GPT-5.6 Sol xhigh configurations. Every point lies to the right of zero; the diamond shows the unweighted mean difference (+15.75 points). Uncertainty intervals are not shown because each configuration has one observation.

GPT-5.5

xhigh

Task	Raw Codex completion	Bello completion	Difference (pp)	Relative change	Raw Codex time	Bello time
Solar	43.78%	53.39%	+9.61	+21.95%	00:16:27	01:29:35
Samtools	20.28%	44.21%	+23.93	+118.00%	00:16:28	02:30:01
Rumdl	46.30%	50.99%	+4.69	+10.13%	00:26:03	03:30:01
Macro mean / total time	36.79%	49.53%	+12.74	+34.64%	00:58:58	07:29:37

Bold completion values indicate the higher observed score within each matched row.

Bello's score was higher on all three tasks. The task-level differences ranged from 4.69 to 23.93 percentage points; the unweighted macro average increased from 36.79% to 49.53%, a gain of 12.74 points (34.64% relative).

Cross-task completion summary

Figure 2. Cross-task completion summary on a common 0–100% scale. Panels (a), (b), and (c) show the matched GPT-5.6 Sol ultra, GPT-5.6 Sol xhigh, and GPT-5.5 xhigh comparisons. The unweighted macro differences are +20.50, +15.75, and +12.74 percentage points, respectively.

Task-level configuration profiles

The following panels compare all three complete three-task configurations: GPT-5.5 xhigh, GPT-5.6 Sol xhigh, and GPT-5.6 Sol ultra. Each panel contains exactly six bars (Raw Codex and Bello for each model–effort setting), ordered by increasing completion score. Bello precedes Raw Codex when scores are tied. Ordering is descriptive and does not imply compute equivalence.

Figure 3a. Solar completion scores for the six model–effort configurations, sorted from lowest to highest. The two formerly tied values are shown at their available precision: Codex GPT-5.6 Sol ultra at 53.13% and Bello GPT-5.5 xhigh at 53.39%.

Figure 3b. Samtools completion scores for the six model–effort configurations, sorted from lowest to highest.

Figure 3c. Rumdl completion scores for the six model–effort configurations, sorted from lowest to highest.

A shorter quality–efficiency balance

We also tested a shorter C+A+C schedule: it reached 63% completion on Samtools and 79% on Rumdl. Each completion-review or adversary pass is designed to find every material defect it can in the solution snapshot it receives, so each successive pass tends to deliver a smaller quality gain at roughly the same per-pass cost. The C+A+C results, together with an intermediate run in which the first two completion reviews delivered roughly 80% of the eventual improvement, support this diminishing-returns pattern. We therefore recommend C+A—one completion review followed by one adversary pass—as the best balance of quality, time, and cost. We expect it to retain about 60–70% of the full schedule's quality gain: against the roughly 35% average relative improvement observed above, that corresponds to an estimated gain of about 20% over Raw Codex.

Time and cost remain limitations. We estimate that C+A takes approximately 2.5× as long as Raw Codex and costs approximately 2.7× as much. The absolute impact is much smaller than those multipliers suggest: in our observed ultra runs, Raw Codex used about 0.2–0.3% of a weekly usage limit on a substantial task, while C+A+C used at most about 1.2%. In economic terms, the share of the available budget matters alongside the relative increase: tripling a negligible expense is less noticeable than a 5% increase in something that already consumes half the budget. We are actively working to reduce both runtime and cost without giving up the quality improvement.

Requirements

• Codex CLI installed and authenticated (Bello drives codex app-server; your Codex account provides the models).
• Python 3.11+ and git.
• macOS or Linux.

Verify your environment at any time with bello doctor.

Install

Option A: Codex plugin (recommended if you work inside Codex):

pipx install bello
codex plugin marketplace add AlexeyKulaev/Bello-codex-marketplace --ref main
codex plugin add bello@bello-marketplace

Then open Codex in your project folder and ask it to run Bello on your task file. The plugin checks for updates and launches the run for you.

Option B: standalone CLI

pipx install bello
bello doctor

Bello checks for updates at startup and offers to install them; run bello update to update explicitly.

Quick start

cd your-project
echo "Build a CLI tool that ..." > task.md
bello --task task.md

That's it. Bello starts the coder, supervises the run, and writes .supervisor/FINAL_REPORT.md when it finishes: status, changed files, validations that were run, and remaining risks.

While a run is active you can type into the terminal; your message is routed to the supervisor, not the coder:

Control	Action
/status	Show task, generation, active turn, pending approvals, health.
/pause / /resume	Pause and resume the autonomous loop.
/restart	Request a supervised restart.
/quit	Write state and exit.
any text	Delivered to the supervisor as an instruction or constraint.

Everything the run does is written to inspectable files under .supervisor/ in your project: PROGRESS.md (what has happened), DECISIONS.md (standing decisions), HANDOFF.md (restart context), events.jsonl (full event stream), and FINAL_REPORT.md (the result).

Run modes

Bello is built for walk-away execution. In both primary modes, the coder works inside a disposable, network-isolated snapshot rather than directly in your live project. A fresh-context runtime supervisor evaluates risky or out-of-sandbox actions, catches drift, and manages recovery; unsupported requests and supervisor failures fail closed. Only an accepted, policy-checked patch is transferred back to the project.

The modes differ in what happens after the coder reports validated readiness.

Everyday (default)

Everyday is for short and medium tasks. A fresh project uses GPT-5.6 Sol at xhigh for both the coder and full runtime supervisor, with Luna handling routine cheap runtime triage. Completion review and the adversary are off.

bello --task task.md

The run finishes once the coder's readiness passes Bello's required validation gates. Runtime supervision remains active throughout the run; only the final review loop is skipped.

Deep Work

Deep Work is for long, demanding tasks with many details and edge cases, where quality takes priority over time and cost. It adds an independent completion reviewer and an adversarial tester, both GPT-5.6 Sol at xhigh by default.

The default Deep Work schedule is C+A:

• 1 completion-review round.
• 1 adversary pass in a disposable snapshot.
• No scheduled post-adversary review rounds.

If completion review returns a defect, the coder fixes it before the adversary runs. If the adversary reports no candidate finding, the run completes. A candidate adversary finding always receives one independent completion-review adjudication so Bello can reject a false positive or return a real defect to the coder; this conditional integrity check is not a scheduled +C phase.

To enable Deep Work, run bello config, set completion-review to true, then set adversary to true. The revealed schedule values default to 1, 1, and 0: one completion-review return budget before one adversary pass, with no scheduled post-adversary review rounds. For a single run without rewriting the saved config:

bello --task task.md --completion-review=true --adversary=true

Custom

For experiments, configure the coder, runtime supervisor, completion reviewer, and adversary independently. Each role can use any available GPT-5.6 variant and its own reasoning effort, and the review and adversary budgets can be combined freely.

Configuration

Open the interactive editor from your project folder:

bello config

It creates and edits .supervisor/config.json. Every value is saved as you press Enter; future runs in this folder use these settings automatically.

The editor starts in Everyday mode for a new project and only shows settings that can affect the selected pipeline. Turning on completion-review reveals the completion reviewer and review budget. Turning on adversary then reveals the adversary model and the complete C+A schedule.

For each visible role, select GPT-5.6 and then choose Sol, Terra, or Luna in the variant row. Sol and Terra support reasoning effort from low through ultra; Luna supports low through max. Active primary roles default to GPT-5.6 Sol at xhigh; cheap runtime triage uses Luna.

CLI flags override their corresponding saved settings for one run and never rewrite the project config. Settings without a CLI flag, including cheap runtime and review budgets, are changed through bello config.

Setting	Default	What it does
task	absent	Default task file for this folder. When set, plain bello runs it; --task always overrides.
coder-mod	GPT-5.6	Model family for the coder thread.
coder-5.6-variant	Sol	GPT-5.6 variant for the coder: Sol, Terra, or Luna.
coder-intelligence	xhigh	Coder reasoning effort, limited by the selected variant.
runtime-mod	GPT-5.6	Model family for fresh-context runtime checks, including risky-action judgment and drift detection.
runtime-5.6-variant	Sol	GPT-5.6 variant for the full runtime supervisor.
runtime-intelligence	xhigh	Full runtime supervisor reasoning effort.
completion-mod	GPT-5.6	Model family for the independent read-only completion reviewer. Hidden in Everyday mode.
completion-5.6-variant	Sol	GPT-5.6 variant for completion review. Hidden in Everyday mode.
completion-intelligence	xhigh	Completion reviewer reasoning effort. Hidden in Everyday mode.
adversary-mod	GPT-5.6	Adversarial tester model family. Visible only when the adversary is enabled.
adversary-5.6-variant	Sol	GPT-5.6 variant for the adversary. Visible only when the adversary is enabled.
adversary-intelligence	xhigh	Adversary reasoning effort. Visible only when the adversary is enabled.
speed	usual	fast uses the Codex Fast service tier for coder, runtime-supervisor, and completion-review turns. Adversary turns are unchanged.
cheap-runtime	true	Let Luna dismiss routine runtime checks before invoking the full runtime supervisor. Human messages, approvals, and mandatory checks bypass triage.
start-over	true	true removes prior Bello logs, archived runs, and recovery data; false preserves them. Both start fresh active state and leave project files unchanged.
completion-review	false	false is Everyday. true enables the independent completion-review loop and reveals its settings.
adversary	false	Enable the adversarial tester before completion. Requires completion review.
max-reviews / max-reviews-before-adversary	1	Completion-return budget. Without an adversary it is shown as max-reviews; with an adversary it limits returns before the first pass. An earlier accept starts the adversary immediately. 0 skips these rounds; Unlimited removes the cap.
max-adversary-runs	1	Maximum adversary passes in Deep Work. 0 disables the adversary.
max-reviews-after-adversary	0	Maximum additional completion-review rounds after each adversary pass. At the limit Bello starts the next pass or completes after the final one. 0 schedules none; Unlimited removes the cap. A candidate adversary finding is still adjudicated once.
clean	false	Warning: deletes everything in the folder except the task file and configured protected paths before starting. Only for disposable folders where you want a from-scratch build.
protected-path	absent	Paths the coder must never write to, such as golden tests, fixtures, or production configs. They are also preserved by clean.

Command reference

bello                 # run the configured task in the current folder
bello --task TASK.md  # run a specific task file
bello config          # open the interactive config editor
bello doctor          # check Python, git, Codex, auth, app-server support
bello update          # update Bello to the latest version
bello update --check --json  # machine-readable update status
bello --version       # installed version, latest version, update status

Run flags (each overrides the saved config for one run):

Flag	Meaning
--task PATH	Task file to run.
--coder-mod M	Coder model.
--runtime-mod M	Runtime supervisor model.
--completion-mod M	Completion reviewer model.
--adversary-mod M	Adversarial tester model.
--coder-intelligence V	Coder reasoning effort.
--runtime-intelligence V	Runtime supervisor reasoning effort.
--completion-intelligence V	Completion reviewer reasoning effort.
--adversary-intelligence V	Adversarial tester reasoning effort.
--fast[=true|false]	Codex Fast service tier.
--start-over[=true|false]	Fresh .supervisor/ state.
--completion-review[=true|false]	Completion-review loop on/off (false = Everyday and disables the adversary).
--adversary[=true|false]	Adversarial tester on/off.
--adversary-runs N	Adversary pass budget; 0 disables.
--clean[=true|false]	Warning: wipe the folder except the task file and protected paths before starting.
--protected-path PATH	Protect a path from writes; repeat for multiple paths.

Environment variables: BELLO_SKIP_UPDATE_CHECK=1 skips the startup update check; BELLO_PROMPTS_FILE=/path/to/prompts.toml points Bello at an alternative prompt file for experiments; BELLO_CONFIG_ANIMATIONS=0 disables motion in the interactive config editor.

License

Bello is released under the MIT License. See LICENSE.

Contributions require signing the project CLA; a bot will prompt you on your first pull request, and you only sign once.