How multiple agents share one identity and one coherent history.
AISHNA, MEMORY, and SOUL as a practical framework for auditable, collaborative multi-agent systems.
In human teams, trust comes from shared values, shared memory, and clear roles. You know what your collaborators stand for. You know what happened before. You know what you're each supposed to do.
Most multi-agent systems skip this. Agents are isolated, amnesiac, and unclear about each other's values or constraints. The result: systems that are hard to trust, hard to debug, and hard to steer.
This framework brings three proven ideas together to fix that:
SOUL defines what each agent (or the whole system) stands for: core values, constraints, operating principles, vocabulary preferences.
It answers: "What are we? What do we care about? What are we not?"
From OpenClaws architecture, refined for auditable systems. Written in human-readable declarations.
MEMORY is an append-only log of decisions, context, and outcomes. Every move is documented. New agents inherit the full history.
It answers: "What happened? Why? Who decided that? What did we learn?"
Agents can query MEMORY to understand prior decisions without starting from scratch. Outputs are auditable because the trail is explicit.
AISHNA (Arabic: "our lived experience," 1st person plural past tense) is the frame where multiple agents share the same SOUL and MEMORY—together.
It answers: "Who are we, together? What is our shared coherence?"
The plurality layer. Multiple agents can operate under one collective identity while maintaining individual roles. Coherence is explicit, not hoped for.
Traditional multi-agent systems treat agents like isolated tools. This framework treats them like team members:
The framework is designed to be simple to implement and human-readable:
A human-readable file declaring the system's identity, values, and constraints. Plain text, version-controlled, discussable.
An append-only narrative log. Each decision, outcome, and lesson is recorded. Indexed for agent access and human review.
Documents the shared state of multiple agents: what they collectively are, what they've collectively done, and collective constraints.
All three are plain-text artifacts. They can be in git, versioned, discussed in pull requests, and inspected by humans at any time.
This framework is not theoretical. It's active in the platform at aishna.b1c3.dev, where AISHNA.md serves as the collective coherent memory for collaborative agents.
The platform demonstrates:
This framework pairs well with SFXL ("simple, familiar, extensible, localized")—a concept exploring how to use everyday words like "windows," "boundaries," and "seasons" to frame system behavior in ways humans naturally understand.
While SFXL focuses on semantic clarity in language choice, AISHNA/MEMORY/SOUL focuses on collective coherence and auditability. They complement each other: SFXL makes systems legible through language; this framework makes them trustworthy through transparency.
This is called "Open Operating Philosophy" because every layer is intentionally transparent:
Teams can fork the framework, adapt it to their constraints, and build on it. Partners can audit it. Communities can learn from it.
Systems built on open operating principles are harder to mislead and easier to improve together.