Polyphonic Decision Engine
Routes requests by task complexity and ambiguity, with an auditable record of each decision.
- Source repo
- naberbabammm34343/llm-task-orchestrator
- Stars
- ★ 124
- Last updated
- today
- Primary language
- HTML
- FA score
- 26/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- Code review teams can route syntax checks, style compliance, and simple bug detection to faster models.
- Not a fit if
- Teams that need documented install commands and a runnable example
- Teams that need confirmed provider, pricing, or endpoint details
- Source review
- 26/100 · Major gaps 5 safety controls not found
What does this agent do, and when should you use it?
Polyphonic Decision Engine is an orchestration layer for routing large language model tasks, directing high-volume, predictable work to faster models and ambiguous or reasoning-intensive work to more analytical models. The README describes three layers: classification produces a task profile with confidence scores, routing applies rules and handles fallbacks, and audit records decisions and responses. It documents API gateway, embedded library, and CLI integration modes, with YAML profiles for routing rules, model endpoints, fallback hierarchies, and classification parameters. It also describes concurrent requests, priority queues, batching, custom routing profiles, and performance analytics. The README does not provide installation commands, runtime requirements, named provider adapters, or a runnable invocation, so implementation and deployment details need confirmation before adoption.
On receipt of a request, the engine classifies it across dimensions such as complexity, ambiguity, novelty, and domain specificity, then selects a model using YAML routing rules and manages the request lifecycle. If the primary model does not produce a satisfactory result, it escalates through the configured fallback hierarchy. It records classification scores, routing decisions, response times, and output quality metrics; audit logs can be exported for analysis or compliance reporting. The README also describes request batching, priority queues, language-aware preprocessing, and a workload analyzer for historical request patterns, but gives no concrete command or API names for these components.
- Code review teams can route syntax checks, style compliance, and simple bug detection to faster models.
- Teams maintaining legacy code can separate mechanical renames from dependency restructuring and other semantic changes.
- Architecture groups can send ambiguous design questions with competing trade-offs to deeper reasoning models.
- Documentation teams can route format-driven API documentation separately from conceptual material requiring original explanation.
How do you install or deploy this agent?
The README gives no install command, package name, source build steps, or service deployment checklist. It describes API gateway, library, and CLI integration modes and says configuration is YAML-based, but does not specify runtime requirements, dependencies, model endpoints, or credential formats. Confirm those details from the implementation before adopting it.
How do you use this agent?
The README says profiles can define classification dimensions, routing tiers, and model names, but does not provide a runnable config path, startup command, API request format, or CLI usage. This is the configuration structure shown in the README; it is not a complete runnable setup:
routing:
dimensions:
- complexity
- ambiguity
- novelty
- domain_specificity
tiers:
high_volume:
complexity_max: 4
ambiguity_max: 3
model: fast_general
deep_reasoning:
complexity_min: 7
ambiguity_min: 6
model: analytical_precisionThe README recommends profiling historical requests, defining routing rules, configuring model endpoints and authentication, testing in sandbox mode, then refining based on audit data. It does not include commands or invocation examples for those steps.
What are this agent's strengths and limitations?
- Classifies requests by complexity, ambiguity, novelty, and domain specificity rather than relying only on keyword matching.
- Records classification scores, routing choices, and response metrics for auditing and tuning.
- Supports configurable fallback tiers, request priorities, and batching for different task paths.
- Documents API gateway, library, and CLI integration modes for different application architectures.
- No install commands, runtime requirements, or complete runnable examples are documented, making deployment effort hard to assess.
- Named model providers, endpoint adapters, and credential formats are absent, so interoperability is unclear.
- Claims of under 100 ms classification overhead and 99.9% routing determinism lack benchmark methods or supporting results.
- License information conflicts: repository metadata says unknown while the README claims MIT.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Polyphonic Decision Engine This agent | 26 · Major gaps | Self-hosted service | ★ 124 | today | HTML | — |
| Kodus | 66 · Some gaps | Hosted serviceFreemium | ★ 1.5k | today | TypeScript | OpenAI API · Claude API |
| Pi Dynamic Workflows | 88 · Good | Agent plugin / skillFree + model costs | ★ 555 | 17d ago | TypeScript | — |
| Pilotfish Multi-Model Orchestration | 81 · Good | Agent plugin / skillFree + model costs | ★ 711 | 14d ago | Python | Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README claims request payloads are not persisted and says users manage their own model credentials, but it does not explain actual data flows, log contents, or credential protection; those scores are therefore only 1. A workflow runs hourly and writes randomized content and commit messages to main with contents: write, with no user confirmation, least privilege rationale, or rollback guidance; external effects and permission controls score 0. The code references actions/github-script@v7, but provides no dependency security review. Unverified publisher identity is not treated as a deduction; repository materials also do not clearly identify an author or maintainer, so source attribution scores 0.
The README describes classification, routing, audit, and fallback features without implementation code or tests to support them. Its determinism and performance claims are not verifiable, so self-consistency scores 1. Only one GitHub Action reference is provided, with no explanation of model endpoint or runtime dependency availability. The workflow retries 409 conflicts and throws other errors, but the materials do not explain Agent request failure messages or recovery; failure messages score 1.
The README gives use cases including code review, legacy refactoring, architecture decisions, and documentation, so audience and scenarios score 2. It mentions routing dimensions and YAML configuration, but gives no substantiated classification rules, threshold definitions, or capability boundaries, and does not clarify when the Agent should be triggered. Environment requirements are also absent, so the other scores are 1.
The README is organized by features, scenarios, architecture, configuration, and getting started, scoring 2; however, Getting Started offers conceptual steps without installation commands or actionable configuration instructions. The only example is a short YAML fragment, and there is no FAQ. The disclaimer acknowledges that output quality is not guaranteed, scoring 1, but does not detail other limitations. The README claims an MIT license and refers to a LICENSE file, which is not included in the provided source material and cannot be verified, scoring 1. There is no version or change history. Support is described generally, but no verifiable maintainer or update path is provided, scoring 0.
If implemented as described, task-based routing could reduce model-selection effort. But the source material includes no actual Agent implementation, usable output example, or evidence of outcomes, so output usability, marginal value, and cost-benefit each score only 1.
Claims about semantic classification, determinism, performance, audit trails, and an analytics dashboard have no supporting code, tests, or other materials, so claim traceability scores 0. Only the README and one automated commit workflow are available; there is no implementation or test material for cross-source corroboration, scoring 0. The disclaimer says performance and quality should be validated for specific workloads, showing limited separation of claims from guarantees and scoring 1; the main feature claims are not otherwise distinguished as implemented, planned, or promotional.
- Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Not found in source: verifiable attributionInstall from the official repo or registry and check the publisher and URL to avoid look-alike packages.
- The workflow automatically writes a file and randomized commit message to main hourly, using contents: write; verify its intended purpose and repository impact before enabling it.
- Core Agent features and performance metrics are described only in the README; the provided materials contain no implementation or test evidence.
- The README claims an MIT license, but the provided materials omit the LICENSE file; installation steps, maintenance responsibility, and update path are also unclear.