# Alignbase

> Self-improving, portable context for AI agents. Host your AGENTS.md, Skills, and memory, then watch as every agent's learnings improve the whole fleet.

## What it is

Alignbase gives teams one place to manage agent context. Engineers use it to host team-wide AGENTS.md, approved Skills, and short-term Memory alongside product context, policy, customer context, and cross-repo context that repo-level files and knowledge bases do not reliably share or govern.

Repo AGENTS.md files should still own repo-local commands, conventions, and codebase-specific rules. Alignbase is for context that applies beyond one repository or needs permissions, reviews, version history, delivery controls, and audit records.

## Context is still in the desktop era

1. **Vendor lock-in.** Teams configure AGENTS.md, Skills, Memory, MCPs, plugins, and other inputs for each agent vendor, then repeat the work when they change tools.
2. **Cold start.** Agents begin without the current priorities, project context, operating rules, Skills, or working Memory they need, so teams pay for the same learning again.
3. **Noncompliance.** Security cannot prove which policies, permissions, project context, Skill versions, and Memory versions reached an agent before it acted.
4. **Misaligned fleet.** Teams cannot coordinate agents across branches, repos, users, models, and harnesses for strategic or tactical work.

## Why AGENTS.md, Skills, Memory, and knowledge bases are different

1. **AGENTS.md can't improve the whole fleet.** Repo-local AGENTS.md files are useful, but teams need one place to update context and improve agent output across users, repos, and branches.
2. **Skills need governed access.** Skills are useful optional context, but teams need control over which agents can read, install, update, and use them.
3. **Memory needs to stay short and current.** Each agent can use one assigned Memory file for working state, user corrections, stable preferences, unfinished work, failed approaches, and known traps. Agents maintain it directly when permitted.
4. **Knowledge bases are too big for working recall.** They are broad and often noisy or out of date. Memory does not replace a knowledge base, and a knowledge base should not be used as an agent's short-term Memory.

## Context that belongs in Alignbase

- Company overview, strategy, and KPIs
- Product description, vision, and ICP
- Security, privacy, and compliance rules agents should not miss
- Cross-repo architecture decisions and active migrations
- Engineering standards and review policy that apply across teams
- Active outages, incidents, and maintenance windows

## Skills that belong in Alignbase

- Testing notes, commands, and fixtures
- Release and rollback procedures
- Design system rules and component examples
- Security review checklists
- Customer-specific implementation notes
- Scripts, templates, and task-specific references

## Memory that belongs in Alignbase

- Current working state
- User corrections and stable preferences
- Unfinished work and next steps
- Failed approaches and known traps

## Context that belongs in repo AGENTS.md files

- Local setup commands
- Test, lint, and build commands for that repository
- Repo-specific code style
- Directory-specific conventions
- Commands and notes that only apply inside one codebase

## Product model

1. **Context, Skills, and Memory repository.** The system of record where approved context and Skills are reviewed and governed, and where assigned Memory stays versioned without a review queue.
2. **Context distribution.** The delivery layer that routes the right context and Skill access to each agent and supplies its one assigned Memory.
3. **Audit and controls.** Records of who changed each input, who can edit it, and which context, Skill, and Memory versions an agent received at a point in time.

## Benefits

1. **Portable, agent-agnostic context & configuration to avoid vendor lock-in.** Define context, AGENTS.md, Skills, Memory, MCPs, and plugins once, then use them across agent vendors.
2. **Agent strategic alignment to fix the cold-start problem and deliver better, faster, cheaper.** Give every agent current priorities, context, rules, Skill access, and working recall.
3. **Agent context governance for compliant, auditable agent inputs.** Control and record what reaches each agent and tie its inputs to approved versions.
4. **Fleet alignment for self-improving, coordinated agent fleets.** Share approved learnings and coordinate agents across branches, repos, users, models, and harnesses.

## FAQs

**What is AI coding agent context?** AI coding agent context is the product, policy, customer, architecture, and team knowledge an agent needs to implement work correctly. Repo-local commands and conventions still belong in repo AGENTS.md files. Alignbase is for team and cross-repo context those files cannot manage well.

**How is Alignbase different from AGENTS.md?** AGENTS.md is useful for repo-local rules, commands, and conventions. Alignbase gives teams a managed AGENTS.md layer with Skills and short-term Memory, plus versions, permissions, and audit history.

**Does Alignbase replace repo-level AGENTS.md files?** No. Repo-level instructions are still useful for local commands and codebase rules. Alignbase adds the team context that lives outside one repo, then routes the right bundle, Skill access, and assigned Memory to each agent. A team can be a squad, function, or whole company.

**How do Skills fit into Alignbase context?** Some context should be loaded when an agent starts. Other context should be available only when the task calls for it. Alignbase manages both, with tags, permissions, versions, and audit records.

**How does Memory fit into Alignbase context?** Memory is short-term working recall, not a knowledge base or published instruction. Each agent can have one Memory file, many agents can share a file, and admins can decide whether each agent may update it.

**What does self-improving, portable context mean?** It means agents can carry working recall across sessions in Memory, while durable learnings can become reviewed, versioned context or Skills that the fleet uses later.

**Why not just point agents at a knowledge base?** Knowledge bases are usually too broad, noisy, and stale for agent startup context. Alignbase separates approved context and Skills from concise, agent-maintained working Memory.

**Which engineering context should go into Alignbase?** Good starting points include product goals, customer constraints, security rules, cross-repo architecture decisions, active migrations, rollout plans, incidents, and review policy that applies beyond one repo.

**What benefits should engineering teams expect?** Alignbase gives teams context governance, aligned agents, fleet-wide improvements, and token savings from building the right thing in fewer iterations. Engineers spend less time repairing missed context because useful updates can become shared context or Skills.

**How do teams start?** Start with one team, service area, or repeated workflow where coding agents miss team-level context. Connect an agent, host the AGENTS.md and Skills it should use, then assign a Memory for working recall across sessions. Monitor agent work and publish durable learnings back to the fleet.

## Facts

- **Product:** Alignbase
- **Company:** Sunpeak AI, Austin, TX
- **Category:** self-improving, portable context for AI agents, team-wide AGENTS.md, AI agent Skills, AI agent Memory, context repository, context distribution, context governance, aligned agents
- **Website:** https://alignbase.ai
- **App:** https://app.alignbase.ai

## Pages

- [Homepage](https://alignbase.ai/): self-improving, portable context for AI agents, team-wide AGENTS.md, Skills, Memory, context governance, aligned agents, fleet-wide improvements, token savings, workflow, FAQs
- [Sign in](https://app.alignbase.ai): Alignbase web app