Skip to content
@Allura-Ecosystem

Allura-Ecosystem

Agentic AI framework, governed memory engine, and agent harness for enterprise software delivery

allura

Allura wordmark

Memory that shows its work.

Allura gives agentic systems a governed way to retain evidence, review what becomes reusable knowledge, and preserve the context behind every retrieval.

Governance is part of the flow: a proposed action passes through policy, human approval, and evidence retention around the Allura agent workflow.

Product-support diagram. Allura’s published capabilities and implementation boundary are documented below.


How the ecosystem works

Allura Ecosystem Map: how memory, teams, and plugins work together around the Allura Memory/Brain core.

Allura is the governed memory layer for AI. It gives teams and agents a trusted memory, connected context, and the tools to turn work into intelligent outcomes. The memory engine sits at the center; Team RAM, Team Durham, Allura Plugins, and Curator & Governance feed it evidence and draw governed context from it, and everything converges on the apps, agents, and outputs the system produces.

Team RAM — AI engineering workforce

Team RAM: AI Engineering Workforce. Task intake, Brooks orchestration, specialist functions (architecture, coding, review, testing, governance, routing), verification, merge and delivery, and lessons and memory.

Team RAM is the engineering workforce inside Allura: architecture, coding, testing, governance, routing, and self-improving workflows. You define the goal; the team builds it, verifies it, and makes it better.

Team Durham — AI design & brand workforce

Team Durham: AI Design and Brand Workforce. Discovery, strategy, messaging, UX, UI, accessibility, visual QA, and analytics loop.

Team Durham is the design, brand, and experience workforce inside the Allura ecosystem: strategy, UX, content, visual design, accessibility, and presentation quality — turning ideas into experiences that are clear, beautiful, inclusive, and on-brand.

One task across the Allura workforce

One task across the Allura workforce: a user request enters, Allura Memory/Brain understands and recalls, Team RAM builds the logic, Team Durham shapes the UX and message, plugins supply tools, Curator and Governance verifies, the final product is delivered, and feedback returns to memory.

Every task flows through the same governed loop: memory understands intent and retrieves context, the workforces build and shape, governance verifies, and feedback strengthens memory for the next task.

Ecosystem infographics. Allura’s published capabilities and implementation boundary are documented below.


Agentic AI framework & harness

Allura is built as a reusable agent framework and harness for enterprise software delivery — the same capabilities that let engineering teams design, compose, test, and operationalize agentic workflows with correctness, composability, governance, and testability at scale.

Capability What Allura provides
Agent orchestration & execution DAG-based process engine with step lifecycle events, checkpoint continuation, and event-sourced replay for deterministic runs.
Memory patterns Append-only episodic ledger plus a curator-gated semantic layer over one PostgreSQL engine, with branchable working memory and tenant-scoped retrieval.
Policy hooks & tool calling Control plane with 12 syscalls; every mutation requires proof-of-intent then policy evaluation before dispatch, through an authenticated MCP gateway with fail-closed tool calling.
Simulator harness & evaluation Deterministic scenario runner with tool simulators, byte-for-byte receipt replay for determinism proof, and eval suites over scenario runs.
SDK, API & CLI design Typed SDK (ESM + CJS), allura run / allura replay CLI, and a canonical HTTP gateway — one developer surface over the governed contracts.
Framework scale & governance 59 forward-only migrations with forced RLS, immutable SHA-bound receipts, and CI evidence lanes that aggregate unit, build, live-PostgreSQL, benchmark, and evaluation artifacts.

Every capability above is code, not copy — each maps to named source paths in FRAMEWORK.md, and the engineering narrative is in the engineering case study.

Start with the source of truth

Allura_Memory is the authoritative public product repository. Its code and documentation define the engine boundary, memory lifecycle, governance rules, supported interfaces, and current implementation status.

For a framework-level view of orchestration, memory patterns, policy hooks, harnesses and evaluation, and SDK design, read FRAMEWORK.md.

How Allura handles memory

Stage What happens Why it matters
Capture Agent activity is retained as append-only episodic evidence in PostgreSQL. The original context remains available for inspection.
Review Candidate memory is scored and governed knowledge is routed through review. Reuse is a considered decision, not an automatic side effect.
Promote Approved knowledge is promoted into versioned PostgreSQL canonical graph tables. The reusable record has a clear state and lineage.
Retrieve Scoped memory is accessed through controlled MCP and API interfaces. Retrieval respects the system’s boundaries.
Account for it Source, actor, decision, and audit context remain attached through the lifecycle. A memory-backed response can show where it came from.

The evidence model in six plain-language visuals — conceptual patterns the ecosystem implements, not a live system.

Source, review, and result stay connected to a durable evidence record.
Every action leaves proof
Two labeled workspaces on either side of a policy boundary; private work stays in its own workspace.
Access has boundaries
A proposal passes a policy check, then a person reviews it before it is approved or queued.
People review important changes
New versions supersede old ones while append-only evidence and traceable change stay in place.
Memory keeps its history
Scenario, run, evidence, and review loop into a confident release.
Test before release
An answer stays attached to its source, approved memory, and decision record.
An answer can show its work

Product surfaces

Surface Role
Allura Memory Engine Canonical memory data plane and source of truth.
MCP and API Primary integration path for agents and applications.
Memory Command Center Optional, launch-gated operator view over the same governed contracts.
Curator and RuVix governance Human review, policy disposition, and evidence receipts.

The Memory Command Center is not a second memory system. It may inspect and request governed actions, but it cannot bypass the engine or own canonical state.

The framework applied: bumblebee

Bumblebee is a governed supply-chain threat-intelligence plugin: a pinned upstream scanner wrapped as a deterministic, policy-gated Allura agent tool. It is the framework above in production form — orchestration, least-privilege credentials, fail-closed ingest, policy-gated promotion, and evidence retained at every step.

From scan to trusted exposure: a developer endpoint is scanned by the pinned Bumblebee scanner, results pass lease issuance, secure ingest, validation, and promotion or hold, and exposure recomputation feeds a governed response back to the developer endpoint.
From scan to trusted exposure
Zero trust by design: pinned scanner, least-privilege credentials, source-bound lease, HTTPS-only ingest, schema and scope checks, and fail-closed processing. The scanner reports. The server verifies.
Zero trust by design
Role-based access and scoped authority: a principal with role and permission, source binding, and a scan lease with a short-lived ingest token reaches an allowed action inside a tenant and workspace scope, or access is blocked.
Role-based access and scoped authority
A vulnerability has to prove itself: a scanner finding becomes a trusted exposure only after the installed package, matching version, and bound advisory catalog are all verified.
A vulnerability has to prove itself
Bad scans never replace good inventory: complete, bound, consistent scans promote into the trusted inventory; partial or conflicting scans are held as evidence only.
Bad scans never replace good inventory
Bumblebee runs on the Allura governance layer: identity, RBAC, and tenant isolation on one side; audit evidence, telemetry, and policy gates on the other.
Bumblebee runs on the governance layer

Product infographics for the bumblebee plugin. The pinned scanner, ingest pipeline, and policy gates behind every claim are in the bumblebee repository.

Principles we build around

  • Evidence before completion claims.
  • Human approval before governed promotion.
  • Provenance before reuse.
  • Tenant boundaries before retrieval.
  • Explicit degraded or unknown states.
  • Reproducible verification for public claims.

The ecosystem

GitHub may display other public repositories owned by this organization. Some are supporting experiments, delivery harnesses, or reserved namespaces; visibility does not make them equal parts of the Allura product.

Repository Role
Allura_Memory Canonical governed memory engine, MCP server, SDK, policy hooks, and evidence authority
allura-team-ram Canonical governed multi-agent software-delivery harness for OpenCode, Claude Code, and Codex
team-durham Canonical governed multi-agent brand-production harness for strategy, design, accessibility, evidence, and QA
bumblebee Governed supply-chain threat-intelligence plugin — a pinned scanner wrapped as a policy-gated Allura agent tool
allura-plugins Installation catalog and pinned generated exports from the standalone canonical repositories
Allura-ecosystem Ecosystem map, governance policies, enforcement plugin, and architecture index

Allura is memory that shows its work. Its public claims should do the same.

Popular repositories Loading

  1. Allura_Memory Allura_Memory Public

    Canonical allura product: a self-hosted governed memory engine for AI systems.

    TypeScript 1

  2. allura-team-ram allura-team-ram Public

    Governed multi-agent software-delivery harness for OpenCode, Claude Code, and Codex

    Python

  3. allura allura Public

    Reserved namespace. The canonical product repository is Allura_Memory.

  4. mortagate mortagate Public

    Microsoft Copilot Cowork skills for governed mortgage evidence review and audit-packet drafting

    Apex

  5. allura-plugins allura-plugins Public

    Installation catalog and pinned generated exports for governed Allura agent teams

    JavaScript

  6. .github .github Public

    Allura-Ecosystem organization profile and community health files

Repositories

Showing 10 of 11 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…