How it works

AI sessions forget.
Projects shouldn't.

Agenties stores your plan, team, memory, and evidence outside the chat — so every session resumes where the last one stopped, on any model, after any restart.

IntentPlanWorkVerifyRemember
See where it breaks first ↓
The problem

Every new session starts from zero.

You re-explain the project. The context window fills before real work starts. An agent says done — and you have no way to check. This is missing infrastructure, not a model problem.

Without Agenties
  • Open a new chat — re-explain the entire project again
  • Context window consumed on catching up, not moving forward
  • No record of decisions made or dead ends already tried
  • Tasks closed on the agent’s word, no evidence ever checked
  • Parallel agents silently overwrite each other’s work
  • Token spend invisible until the bill lands
With Agenties
  • Session opens with a compact brief — project context already loaded
  • Token budget spent on progress, not re-orientation
  • Goals, decisions and completed work persist across every restart
  • Issues close only when evidence criteria are satisfied
  • File claims prevent overwrite conflicts before they happen
  • Live token spend and agent budgets on screen at all times
The operating loop

How the work keeps moving

Five stages, one closed cycle. Intent enters as text — evidence exits as proof. Memory feeds the next loop so the project never starts cold again.

IntentYou describe the outcomeOne sentence of intent — no prompt-crafting, no babysitting."migrate all docs to the new v3 format"
PlanIt becomes a tracked goalThe orchestrator breaks intent into a goal with linked, ordered issues.Goal CL-42 · 7 issues · assigned by role
WorkA persistent team deliversScout, builder and reviewer claim files and report back.Scout: 23 files · Builder: migrating · Reviewer: on it
VerifyNothing closes without proof"Done" arrives as evidence — git status, tests, ACs all satisfied.23/23 migrated · tests pass · PR clean ✓
RememberContext carries forwardGoals, issues and evidence persist in local files — the next session re-enters informed, not blank.Goal closed · journal updated · session ready

Memory feeds the next intent — the loop never resets.

What a session actually looks like

Intent enters as text. The orchestrator assembles context from local files, delegates to the right agents, and stores the result — so the next session continues instead of restarting.

Agenties
agenties
Work
Chat
Dashboard
Manager
Monitor
Heartbeats3
Agents
Issues2
Activity
History
Configure
Routines
Team
Settings
Team online0 active
No agents running
DI
disespubli
Free plan
Orchestrator ready
checkout-work
Files
What's left on the checkout goal?
O
Orchestrator· 09:41
Goal CK-1 has 3 issues. CK-1A done. CK-1B (payment form) in progress — Builder resumed last session. CK-1C (order email) still open.
Tell the orchestrator what to do…
claude-sonnet-4-5
Effort: Off
Auto-routes to builder, scout, coordinator, or runs directly⌘+Enter
Everything included

Built for the three things that break AI work

Context lost between sessions. Work closed on a promise. Costs invisible until it’s too late. Every capability is active out of the box — nothing to configure.

Stop re-briefing from scratch
Goals, decisions and team roles live in local files. Close the chat, restart the machine, switch models — every session opens already knowing the project.
  • OrchestratorYou describe the outcome. The orchestrator delegates only when justified, sequences the work, and keeps you informed — no juggling agents yourself.
  • Plans, goals & issuesIntent becomes a goal. Goals break into ordered issues. Issues close only on evidence. Work stays coherent across every session — no scattered prompts.
  • Memory & re-entryGoals, decisions and completed issues live in local files. A fresh session reads a compact brief — not a full chat log — and starts already knowing the project.
  • Local project filesState lives as plain files in your project directory. No cloud dependency, no vendor lock-in. Offline-ready and entirely portable.
Get real evidence, not just claims
A named team picks up each issue. Agents claim the files they need, can't mark done without satisfying the acceptance criteria, and don't overwrite each other.
  • Persistent teamNamed specialists — Scout, Builder, Reviewer — each with a role and a history. They pick up where they left off without being re-briefed.
  • Completion evidenceGit status, test results, acceptance criteria — an issue closes only when evidence criteria are met. Not when the agent feels confident.
  • Heartbeats & routinesSchedule recurring tasks once. Heartbeats fire on a cron — code reviews, health checks, daily digests — whether you're watching or not.
  • Cost & contextLive token spend and active agent budgets stay on screen. See what a session costs in real time — not after it's already burned.
Control from any client, drop whenever
100+ MCP tools expose every orchestrator action. Claude Desktop, Codex, Cursor, or any compatible client drives it — no app to install, no lock-in.
  • MCP control surface100+ tools expose the full orchestrator. Claude Desktop, Codex, or any MCP-compatible client drives Agenties from where you already work.
  • File claims & auditAgents lock the files they need before editing. No silent overwrites, no conflicts from concurrent work. Every action logged — nothing off the record.
  • Skills & hooksWhat works gets saved as a skill. Project-specific behaviors repeat automatically — good patterns don't need to be re-taught each session.
  • IntegrationsWordPress, GitHub, Supabase, Obsidian, NotebookLM, Google Workspace and more. Preconfigured MCP connectors — just add credentials, no wiring.
MCP Control Surface

Works from any MCP-compatible client

100+ MCP tools expose the full orchestrator. Spawn agents, query memory, create goals, close issues, read the audit log — all from whichever AI client you already have open. Claude Desktop, Codex, Cursor, and Claude.ai are tested examples, not the limit.

Claude DesktopClaude.aiCodexCursorAny MCP client
Your AI client
Agenties orchestrator
Goals · Issues · Memory · Agents · Sessions
What’s open in the current sprint?→ list_issues · status: open · goal: active
Spawn a reviewer for the auth PR→ spawn_agent · role: reviewer · context: PR diff
What did we decide about the DB schema?→ query_memory · tag: decision · keyword: schema
spawn_agentlist_issuesquery_memoryget_shared_statecreate_goalsend_messageclose_goal_with_issuesread_messages
Pricing

Free to start. Pro when you need more.

Install once, bring your own Claude Code plan. The orchestrator, memory, persistent team and integrations are all in the free tier. Pro adds early access and unlimited scope — no extra AI billing ever.

Free
0
Forever · no credit card required

  • 1 company, 2 linked projects
  • Full orchestrator + persistent team
  • All integrations (just add credentials)
  • Local memory, goals & issues
  • Heartbeats & routines
  • MCP control surface
Install now
Coming soonTeam / AgencyComing soon
Reach out for early team access

  • Everything in Pro
  • Members, roles & permissions
  • Shared vault & notebooks
  • SSO support
  • Full audit log
Talk to us →

Agenties does not include Claude or Anthropic tokens — AI usage runs through your own Claude Code or Anthropic plan. Pro and Team pricing will be announced to the waitlist first. Joining locks your early-access rate.

Get started

One command away.

Install globally, open any project directory, and write your first intent. Works with your existing Claude Code plan — no extra API key, no extra billing, no cloud account required.

>npm install -g agenties
Read the docs ↗

Connect your stack — preconfigured, just add credentials

WordPressGitHubSupabaseInsForgeObsidianNotebookLMGraphifyGoogle WorkspaceBetaWordPressGitHubSupabaseInsForgeObsidianNotebookLMGraphifyGoogle WorkspaceBeta