Agentic Systems | Zhang Handuo's Site

FrontierAgent

Tue, 25 Aug 2026 00:00:00 +0000

I initiated and lead FrontierAgent, an open-source agent runtime and research workbench for long-running research and file-based tasks. It combines a terminal interface, sandbox tools, and evaluation workflows for single-agent ReAct and Agent Team execution, and shares reusable infrastructure with ApodexHarness through AgentCore.

Demo

A 1-minute 48-second walkthrough of Agent Team in FrontierAgent’s terminal interface.

The problem: keeping a long task coherent

A research task may involve several evidence sources, analysis scripts, intermediate files, and parallel investigations. Simply adding more agents does not settle who owns the plan, when work is complete, or which files constitute the final result. FrontierAgent makes those responsibilities explicit in the runtime.

Its public interface offers two workflows: ReAct for a stateful single-agent loop, and Agent Team for a coordinator that delegates independent assignments and integrates reports. The same workflow engine also supports benchmark execution, connecting interactive use with repeatable evaluation.

Conceptual Agent Team workflow. Specialist roles illustrate possible task decomposition; the coordinator organizes the team for the task rather than using a fixed roster.

Coordination, state, and delivery

The architecture below separates three responsibilities that become tightly coupled during a long run:

Coordination: the main agent maintains an external task board and uses an Agent Bus to dispatch work and collect reports. Execution status and the coordinator’s judgment that a task is resolved are distinct pieces of state.
Execution: sub-agents produce candidate artifacts in working directories. The isolation backend determines whether those directories are private worktrees or a shared mount. Read-only inputs remain separate from mutable working files.
Delivery: a single-publisher lease and an explicit output manifest govern final publication. Reconciliation against the initial output state prevents stale or empty files from being mistaken for newly completed deliverables. Shared scratch space remains outside the collected final output.

Runtime responsibilities across coordination, execution, and publication. The diagram distinguishes working state from final delivery and shows where user follow-ups enter the active run.

These mechanisms address state loss, ambiguous ownership, concurrent output writes, and premature completion claims. They do not by themselves establish that a scientific method or conclusion is correct; task-specific verification remains necessary. The runtime contract is also described in the , Section 3.3 and Table 2.

An inspectable workbench and evaluation path

The terminal product exposes the task board, tool activity, artifacts, and local traces. Checkpoints support resuming a session, and follow-up instructions can steer an active run. The evaluation runner supports artifact collection and rerunning failures, making it possible to investigate where a task broke down rather than only record its final score.

The repository also contains , which evaluates complete scientific deliverables. See the for its task design and reported completion gaps, and for model-and-system results using ReAct and Agent Team. Those benchmark gains reflect the complete evaluated configurations, not a standalone runtime speedup claim.

Explore: · .

Apodex 1.1: Scaling Agentic Intelligence for Complex Work

Mon, 24 Aug 2026 00:00:00 +0000

A useful research agent must do more than retrieve information and produce a report. It needs to inspect real files, execute analyses, maintain state, recover from failure, and leave behind work that can be checked and reused. Apodex 1.1 calls this working capability: sustained, verifiable progress toward a real objective.

I contributed to the agent harness and its evaluation and release infrastructure. My work connects long-running execution, multi-agent coordination, and artifact verification through . The results below describe the complete team-built model and system, rather than an isolated contribution from any one component. The author list on this page is abbreviated; the report contains the full credits.

Two complementary ways to scale agent capability

Environment Scaling expands the file, search, and code environments in which agents can act and receive verifiable feedback. Tasks require the model to work with data and executable tools, rather than rely on text-only answers.

Agentic Coordination Scaling trains task decomposition, delegation, integration of asynchronous results, and replanning. A coordinator can assign independent work to sub-agents, absorb their findings, and organize further verification as the task evolves.

Both depend on a shared runtime. AgentOS maintains task state outside the conversation history, coordinates tool and agent execution, and tracks which files are working material and which are final deliverables. The makes ReAct and Agent Team workflows available through a terminal interface and an evaluation harness.

Reported results: ReAct and Agent Team

The August 2026 technical report compares a single-agent ReAct setting with Agent Team for the same Apodex 1.1 model. Selected results are reproduced below for readability.

Evaluation	ReAct	Agent Team	Difference
APEX-Agents	34.4	38.5	+4.1
GDPVal win rate (%)	69.5	78.8	+9.3 pp
FrontierFinance	48.7	54.3	+5.6
FrontierScience-Research	55.0	63.3	+8.3
BioMysteryBench, Human-difficult	23.5	35.3	+11.8
Humanity’s Last Exam	53.2	56.1	+2.9

Differences are in each benchmark’s reported score units; the GDPVal change is in percentage points. Agent Team uses additional coordinated computation, so these comparisons are not matched-compute estimates of the effect of coordination alone. They are a dated snapshot of the report’s evaluation, not claims about a current leaderboard.

Reported results across professional work, finance, scientific research, and general reasoning. Source: Apodex 1.1 technical report, Tables 3–4; reference-system protocols are described in the report.

From runtime behavior to usable research tools

For researchers, the runtime supports keeping a task moving through changing evidence and user feedback: intermediate files survive, the plan can be revised, and results from parallel branches can be integrated. The official explains how this appears in the workbench through live intervention, visible progress, and review of important claims.

The report also introduces a 35B-parameter Mini model as a locally deployable route to these workflows. This is separate from the full-model results above.

Sources: , Sections 3–4 and Tables 3–5; ; .

Prax / AgentOS and ApodexHarness

Tue, 31 Mar 2026 00:00:00 +0000

I joined Apodex in late March 2026 as one of three founding engineers of Prax / AgentOS and became the technical owner of ApodexHarness infrastructure. The project develops an execution system for complex, long-running tasks.

I built heavy_mode, which combines parallel sub-agents, cross-branch synthesis, global verification, and continuation after interruption. I proposed planner-generated, schema-constrained workflow DAGs and replaced a LangGraph dependency with a compact execution engine supporting branches and verification-retry loops.

I also developed token-budgeted context compression, file-based research memory, tool-specific result budgets, and evidence checks for unsupported numbers and citations. Reliability work included durable agent messages, sandbox isolation, tool permissions, and cancellation-safe event streaming.

The harness connects workflows to a unified evaluation interface. I supported the Apodex 1.1 and 1.2 model releases and ran 240 APEX Agents 1.1 tasks end to end to validate the evaluation path.