Σ‑FDL CodeAgents — Semantic, Explainable Agents

FDL CodeAgents: Python agents that reason via Thesis–Antithesis–Synthesis (Formal‑Dialectical Logic) on top of the OpenAI API, producing transparent decisions and concise, actionable responses.

Comment

Inspiration

Civic and research teams drown in long texts and inconsistent decisions. We wanted an explainable agent that makes its reasoning explicit (FDL: Thesis→Antithesis→Synthesis), reduces noise, and outputs concise, reproducible actions.

What it does

FDL‑CodeAgents provide a lightweight, modular agent core that: accepts prompts and documents; retrieves supporting context (via the OpenAI API wrappers in repo); reasons with Formal‑Dialectical Logic (FDL) into a visible T‑A‑S trace; emits an answer or a compact action plan, suitable for downstream tools. Focus: correctness, semantic density, and auditability of the reasoning steps.

How we built it

Language: Python 3.x Core modules: fdl_compiler.py (FDL logic), agent_core.py (agent loop), api_wrapper.py (OpenAI API access), agent_memory.py, agent_registry.py. CLI & examples: cli_agent.py, examples/ (reference scripts and prompts), benchmark_tests/. Design: clear separation of FDL reasoning from IO; small surface area for plugging alternative model providers later.

Challenges we ran into

Balancing semantic compression with readability of the FDL trace. Guarding against over‑generalization when inputs are noisy or contradictory. Keeping the core provider‑agnostic while relying on a single API in early versions.

Accomplishments that we're proud of

A working FDL engine (Thesis→Antithesis→Synthesis) with compact, readable traces. Minimal agent loop that is easy to audit and extend. Clean repository: examples, and benchmarks for quick validation.

What we learned

FDL makes decisions traceable without bloating tokens.Small, explicit reasoning operators outperform ad‑hoc prompt chains for stability. A thin API wrapper simplifies migration between providers.

What's next for Σ‑FDL CodeAgents Semantic, Explainable Agents

Connectors: optional Notion/Sheet0/email/webhook adapters (separate module) planned. Datastore: a tidb/ experimental branch for hybrid retrieval (Vector + FTS) planned. UI: minimal FastAPI web panel for non‑CLI users planned. Bench: richer stress‑tests on noisy, multilingual corpora.

Built With

  • examples/
  • languages:-python-3.x-ai-/-apis:-openai-api-via-api-wrapper.py-reasoning:-formal?dialectical-logic-?-fdl-compiler.py-(t?a?s)-cli-/-examples:-cli-agent.py
Share this project:

Updates