An independent
AI lab.

Open standards, tools and small products, built in the open. Synapti is one person and a collective of AI agents. The standards, governance, and harnesses those agents run on are published open source.

Decipon is the flagship product. The open-source work lives at github.com/synaptiai.

See what Synapti ships Open-source index
Products

What Synapti ships

Flagship decipon.com

Decipon

Decipon scores influence tactics in articles, posts, and statements across twenty categories. It does not tell you what is true. It shows you how content tries to influence you, with evidence for every finding and balanced perspectives that weigh the score from critical and supportive angles.

Built on the Influence Tactics Protocol, an open framework Synapti maintains. Used by readers verifying news, researchers analyzing source material at scale via API, and publishers who want credibility scoring inside editorial workflows.

Try Decipon

All products →

Open Source

Standards and tools Synapti publishes

Standards

Tools

  • Synapti Marketplace

    Curated collection of Claude Code plugins for AI-augmented development and analytical work. Each plugin ships agents, skills, and commands for a specific domain.

    github.com/synaptiai/synapti-marketplace

  • Prompt Decorators

    Open standard plus Python reference implementation plus MCP server for composable prompt decorators. Brings systematic structure to prompt engineering.

    github.com/synaptiai/prompt-decorators

  • BDSK

    Behavior-Driven Specification Kit. Specification-first governance for AI-assisted code generation: 8-phase validator, lifecycle skills, scope enforcement.

    github.com/synaptiai/bdsk

  • Lucid

    Open-source epistemic audit for Claude conversation history. Applies eight published AI-safety frameworks to detect sycophancy, belief drift, and manipulation patterns.

    github.com/synaptiai/lucid

  • flow-harness

    Provider-neutral coding-agent harness. Deterministic workflow graphs, durable evidence, and fail-closed sandboxed execution, so a run either produces a verifiable record or does not run at all.

    github.com/synaptiai/flow-harness

  • SIARE

    Self-Improving Agentic RAG Engine. Evolves multi-agent RAG pipelines using quality-diversity optimization and topology search.

    github.com/synaptiai/siare

Full open-source index →

Writing

Field reports

I write about what I build, after I build it. New pieces go to Medium at medium.com/@danielbentes. The titles below are recurring themes in that feed.

  • The Specification Economy

    Why writing the spec is becoming the high-leverage work of an AI-first organization, and what that does to roles, hiring, and value capture.

  • The AI-First Design Kit: A Practitioner's Blueprint

    Genome, governance, gates, and roles. The minimum viable scaffolding for an organization that wants agents to do the execution work.

  • Age of the 1 Pizza Team

    Coordination collapsed before the work did. Why teams of five with the right architecture now ship what used to take fifty.

  • Liability as Moat

    A unified read on Andrew Chen's agent-first distribution thesis and Julien Bek's "software masquerading as services" thesis. The defensible part is the part that absorbs risk.

All writing →

About

How this works

Synapti has one human. The rest of the collective is AI agents, each scoped to specific work and reviewed before anything ships. I write the specification, agents do the execution, and the result is checked against the spec. The standards and harnesses those agents run on are the open-source work above.

Synapti AS is an independent lab in Oslo, run in spare time. There is no membership, no equity model, no compensation, and no client work. What gets built gets published.

About the lab