Aira

The Product Development Platformfor engineers and AI agents.

One living memory for the whole team. Every decision preserved. Every plan verified. Always coherent with your codebase.

Explore the Project Ledger →
The closed loop
app.aira.pro/roadmap · global-checkout

Roadmap

ListGanttGraph
Ask Aira
Product Requirements DocumentAvailable
Multi-Region Instant Settlement & Idempotent Webhooks · Grounded in 14 sources
PRD Quality: Advanced · Passed
Scoping96%
Solution Design94%
Operations91%
Overall94%
1F-1Deterministic Idempotency Fence
in_progress96
2F-2Webhook Replay Ledger & Audit Schema
approved91
3F-3Multi-Region Failover OpenTelemetry Spans
approved84
Sprint #15Ledger synced

Measured on our own repositories, October 2026. Aira is built with Aira, and its agents ship through AWF.

pull requests written by AI agents and merged through AWF
1,700+
review threads addressed by agents before merge
31,000+
CI runs across the Aira and AWF repositories
78,000+
Continuous verification across human & AI agent workflows
Zero Drift

Native across your repositories, communication, & coding agents

GitHubBitbucketClaude CodeCodexCursorSlack

Software is no longer built by humans alone.Today, your engineers and your AI coding agents build side by side.Without shared memory, speed creates chaos. With Aira, speed creates coherence.

Memory, planning, agents and execution in one platform.

Each pillar is useful on its own. Together they close the loop from intent to shipped code.

Pillar I · Project Ledger

One memory for every mind on your project.

Decisions, specs, commits and reviews become one source of truth that people and agents both read. Guardian keeps it honest, checking every plan against your live codebase.

Inspect the Project Ledger →
Knowledge Ledger Sync · global-checkout
Synced to 8f3a1c
1F-1Deterministic Idempotency Fence
approved

Standardize checkout idempotency keys across mobile & webhook retries; deprecate legacy Redis lock.

~/.aira/context/global-checkout/entities/checkout.mdaira://context-graph
feature: F-1 · approved (12h scoped)
rationale: prevent duplicate captures under webhook replay
evidence: T-181 · PR #418 · #payments-arch
Reading the exact same memory via disk & Hosted MCP
Staff Eng
Cody · AI
Claude Code
Codex

Pillar II · Intelligent Project Orchestration

From intent to a plan your team can run.

Living PRDs, a roadmap, and sprints sized to real capacity. Aira assigns work across engineers and agents, and flags risk before a deadline slips.

Explore orchestration →
Orchestration · global-checkout
  1. Intent
  2. PRD
  3. Plan
  4. Assign
  5. Risk
Intent · Maya, Product

“Checkout has to work in the EU and the US before Black Friday.”

Living PRD · v1.2Queued
  • Idempotent payments across EU and US regions
  • Session lifetime aligned in both regions
  • Every webhook replay audited
Sprint #15 · sized to capacity0h / 24h
T-181Checkout idempotency fence
4h
T-182Webhook replay migration
5h
T-183Session lifetime middleware
3h
T-184Regional failover tracing
4h
T-185Replay end-to-end suite
3h
Risk flagged: T-184 waits on T-182’s migration, so Aira moved T-182 to day one.

Pillar III · AI Agents

Specialist agents working inside your project.

Nine specialist agents plan, analyze, assign, report and verify. AI team members pick up tasks the way engineers do, and every action is costed and auditable.

Meet the agents →
Team · AI Contributors
OverviewLogsChat
Cody
AI AGENT
Hermes · AWF Operator
Online
Iris-QA
AI AGENT
Hermes · Tier-2 Gate
Online
Swarm Activity · AWF HandoffOpenShell mTLS
CodyRead aira://context-graph/global-checkout/summary.md
CodyMoved T-181 todo → in_progress · locked src/checkout/**
CodySpawned AWF workspace ws_551d0e (claude_code)
Iris-QASpawned AWF workspace ws_08ac1c (codex · regression)
CodyPR #525 opened: [AIRA-T181] Idempotency fence → review
Iris-QAPR #526 validated vs fresh main (142/142 passing)
Actions
0
PRs Opened
0
Compute Cost
$0.00

Pillar IV · Agent Workspace Fabric (AWF)

Orchestrate an entire fleet of agents. Without collisions.

Approved specs become isolated workspaces. Five or fifty coding agents work in parallel in clean Git worktrees, resolve review feedback, and merge through one governed queue.

Explore Agent Workspace Fabric →
AWF ConsoleDocker isolated · owned-path locks
● API: okStream: live
ACTIVE
4
RUNNING
1
MONITORING PR
2
COMPLETED
1,741
[AIRA-T181] Checkout idempotency fence
ws_551d0e·claude_code · sonnet-4-6
PR #525 ↗◆ monitoring_pr
[AIRA-T182] Webhook replay migration 048
ws_08ac1c·codex · gpt-5.5 · xhigh
serialized● running
[AIRA-T184] Failover OpenTelemetry spans
ws_c0f024·cursor · sonnet-4-6
PR #526 ↗◆ monitoring_pr
[AIRA-T179] Ledger evidence snapshot hook
ws_b31e2b·claude_code · sonnet-4-6
PR #521 ↗✓ completed
Merge Queue · tier2 validated vs fresh main● main is green

Other tools remember, plan, or run agents. Aira closes the loop between them.

Plan, build, verify, remember. Every pass starts the next one with more context.

  1. 01 · Plan

    Orchestration turns intent into a plan sized to your real team.

    Orchestration
  2. 02 · Build

    AI agents build it in parallel, each in an isolated AWF workspace.

    AI Agents + AWF
  3. 03 · Verify

    AWF validates every pull request before merge, and Guardian flags drift from the Ledger.

    AWF + Guardian
  4. 04 · Remember

    The Ledger records what shipped and why, with the evidence.

    Project Ledger
What shipped and why feeds the next plan, and the loop starts again.

See Aira in action

A 12-minute walkthrough: one project, from first idea to verified multi-agent execution.

Dmitri Lihhatsov
“We built Aira because the era of managing software through disconnected tickets and status meetings is over. When humans and AI agents share a single, verifiable memory of what was built, why it was built, and what comes next, product development becomes effortless.”
Dmitri LihhatsovCreator of Aira9+ years Product Leadership at Revolut (80M+ users)

Deterministic architecture under the hood.

A governance layer, never a black-box autopilot. Every claim links to evidence, and every action is auditable.

~/.aira/context/<your-project>/
read-only · materialized plain text
~/.aira/context/<your-project>/
├── summary.md                 # project overview, refreshed live
├── entities/                  # people, services, decisions
│   └── <name>.md
├── contradictions/
│   ├── open/                  # things that don't add up yet
│   └── resolved/              # how we sorted them
├── evidence/
│   └── <YYYY-MM-DD>/          # the proof behind every claim
└── timeline.md                # what happened, in order
The Project Ledger
Plain-text context in ~/.aira/context/, readable by Claude Code, Cursor, Codex and any MCP client.
Guardian Verification
Checks specs and plans against commits and PRs, and isolates contradictions before work starts.
Agent Workspace Fabric (AWF)
Per-task Git worktrees and Docker stacks, with a governed PR loop proven across 1,700+ agent PRs and 78,000+ CI runs.

Integrates with your existing stack during transition — GitHub, Bitbucket, Slack, Jira, and Linear.

Explore the Technical Architecture DocsView AWF on GitHub (Apache 2.0)
Inspect the Project Ledger

Questions & architecture

How does Aira fit into our existing engineering stack?
Aira is the unified product development platform and shared memory layer across your repositories, communication tools, and AI coding agents. It works standalone with built-in roadmaps, sprints, and boards—or syncs seamlessly with GitHub, Bitbucket, Slack, Jira, and Linear so your organization can transition without friction.
How do AI coding agents (like Claude Code or Codex) use Aira?
Through Aira's Project Ledger and MCP integration, your project's live state, architectural decisions, and open contradictions are materialized into structured plain-text files. Every human engineer and AI agent reads the exact same verified context, while AWF (Agent Workspace Fabric) runs parallel coding agents in isolated workspaces with automated PR verification.
What if an AI agent gets something wrong?
Aira is a governed intelligence layer, not an unverified autopilot. AWF validates every agent pull request against fresh main and works through review feedback before it merges. Guardian compares what the Ledger says with what GitHub reports and flags drift for your team. Every plan, spec and pull request links back to its source evidence, and corrections go back into the Project Ledger so the next plan starts from the truth.
Is our source code and project data secure?
To analyze your project, Aira stores copies of your repository's source and config files (the first 8,000 characters of each) along with search embeddings, encrypted at rest and in transit. Aira never trains AI models on your data. We are building toward SOC 2 compliance.
How long does onboarding take?
Under 5 minutes. Connect your GitHub repo and Slack workspace via OAuth, invite your team, and Aira starts learning your project immediately. No migration, no data import, no configuration marathon.

Close the loop between intent and code.

Bring your first project in. Onboarding takes under five minutes.

Work email verification · Zero spam, everAlready have access? Log in to Console →