A rough Telegram brief arrives with context, urgency, and sometimes multiple languages in one message.
pam --wave-dispatch
Your AI Workforce, Orchestrated From Your Phone
PhoneAgents is the mobile-first operating layer for AI agent orchestration. Brief, dispatch, monitor, approve, and verify agent waves—without sitting at a desk.
Voicebriefs become coordinated waves
PAMdispatches parallel agent streams
Mobileapprovals and control, anywhere
Vaultevery wave improves the next
PAM Orchestrator
Not one AI tool. A coordinated team.
PAM is the planning and allocation layer for PhoneAgents: it dispatches specialist agent streams, tracks approvals, monitors progress, and keeps work moving across model providers without losing operational context.
PAM extracts intent, turns ambiguity into assumptions, and writes the docs or triage note the agents need.
Specialists launch with separate ownership for code, review, screenshots, docs, or operations.
Builds, tests, visual checks, PR comments, CI, and approval gates are collected before handoff.
The operator gets a concise report: what changed, what passed, what remains, and what is ready for main.
The problem
AI work is powerful, but the operator is still chained to a desk.
The bottleneck is no longer the model. It is the operator—buried under permission prompts, split across open terminals, and unable to step away without losing track of the work.
47
Permission promptsApprove file reads, shell commands, PR checks, and deploy gates one by one.
blocked09
Open terminalsEach agent has its own context, branch, logs, and failure mode.
watching12
Lost decisionsGood fixes disappear into chat history instead of becoming reusable memory.
forgotten05
Parallel streamsPromising in theory, brittle without orchestration and human gates.
riskyCI
Late verificationBuilds, screenshots, reviews, and broken links arrive after attention has moved on.
delayedRB
Vertical productsCustomer workflows need tenant boundaries, audit, and cost controls before scale.
separatePermission prompts in a single wave
Every approval pulls the operator back to the terminal instead of enabling real review.
Parallel output needs ownership
More output only compounds when agents have clear scope, acceptance checks, and visible recovery paths.
Shared memory across sessions
Every agent starts cold when decisions, failures, preferences, and review notes stay trapped in old chat history.
Human still owns the final gate
Pushes, merges, deploys, and destructive changes need operator control without trapping that operator at a desk.
PhoneAgents closes that gap. A conductor that dispatches work, keeps evidence visible, routes approvals to mobile, and turns each completed wave into better context for the next.
The solution
One operating layer. Any model. Any harness. Complete mobile control.
PhoneAgents is not a coding assistant. It is the orchestration layer above your entire agent stack: Telegram, PAM, Memory Vault, skills, and approval policy—all controlled from your phone.
Telegram and mobile cockpit
Voice briefs, approval buttons, tool alerts, screenshots, agent status, and quick actions stay available from the phone.
PAM wave orchestration
Rough intent becomes a stream matrix with ownership, dependencies, acceptance checks, approval gates, and recovery routes.
Harness freedom
Codex, Claude, Hermes, Pi, Playwright, GitHub, local scripts, and future agents can all be coordinated behind one operating loop.
Provider freedom
OpenAI, Anthropic, Gemini, local models, and managed cloud model endpoints are treated as routable capacity, not a lock-in decision.
The vault compounds
Handoffs, ADRs, reviews, preferences, failures, and reusable skills are saved as durable context for the next wave.
Approval policy
Sensitive actions keep a human gate, while low-risk status, screenshots, and verification continue without terminal babysitting.
Parallel waves
Multiple agents can run in separate streams with visible ownership, status, blockers, and verification evidence.
Verification loop
Tests, visual QA, review comments, CI state, screenshots, and changed files become part of the default finish.
Reusable skills
Marketing, coding, research, review, and operations patterns become installable playbooks instead of one-off prompts.
Chained to the desk
One long session, one fragile context window, one operator watching terminals and approving every tool call manually.
Launch five agents from the beach
The operator gives intent, PAM dispatches streams, approvals arrive on mobile, and verification closes the wave with evidence.
One operator system
PhoneAgents is the conductor layer above agents, repos, approvals, and memory.
A brief arrives in Telegram or the Mini App, PAM turns it into coordinated streams, agent execution handles the repo work, commercial assistants become proof lanes, and the Memory Vault ensures every agent starts warmer than the last.
Control loop01
Raw input becomes usable scope.
Voice notes, mixed-language context, screenshots, and rough instructions are normalized into docs, triage, tasks, and constraints.
Execution02
PAM dispatches isolated streams.
The Planning and Allocation Manager maps ownership, launches specialist agents, watches blockers, and keeps risky writes behind human gates.
Memory03
Verification becomes the default finish.
Agent work ends in tests, visual QA, PR review, merge ledger, and a report that can be reused by the next wave.
Memory Vault
compounding context
compounding context
Agent OpsAI dev cockpit
Vertical appsseparate product decks
PAMwave orchestration
Skillsreusable agent playbooks
Telegramapprovals and alerts
Future appssame memory layer
01
Products write memory.
Agent runs, marketing assistants, sales workflows, vertical apps, and skills produce real operational data: code changes, reviews, campaigns, failures, and fixes.
02
The vault normalizes the learning.
Memory files, skills, ADRs, and handoffs become searchable context that future agents can load before touching a task.
03
PAM routes smarter waves.
Each new wave starts with fewer cold starts, clearer constraints, better defaults, and explicit approval boundaries.
04
The ecosystem compounds.
More apps create more context. More context improves agents. Better agents ship better apps.
Memory advantage
Every agent remembers everything that matters.
The Memory Vault is the second brain behind PhoneAgents. Sessions, decisions, mistakes, screenshots, review notes, preference memories, and reusable skills are saved into a Git-versioned knowledge base so the next agent starts warmer than the last.
BriefVoice, docs, context, constraints.
WavePAM dispatches bounded streams.
VerifyTests, screenshots, review, CI.
IngestHandoff, ADRs, lessons, skills.
ImproveThe next run loads better defaults.
Memory Vault
PAM waves
Agent PRs
Skills
Architecture reviews
Traction and velocity
Real waves. Verified artifacts. Already running.
PhoneAgents ships through its own operating loop. Every metric below is real output—not a wireframe, not a demo account.
$0 developer salaries paid while the system shipped real, verified artifacts.
Cloud-ready primitives
The control plane is built for secure, observable, scalable agent work.
PhoneAgents has the primitives enterprise teams need: runtime isolation, gateway policy, memory, identity, observability, audit trails, cost controls, and provider routing. The public site keeps the story platform-first; partner-specific architecture detail can live in follow-up material.
Bounded agent execution
- Separate worktrees, file ownership, tool policies, and approval gates.
- Clear path to containerized workers and tenant-specific execution boundaries.
Vault to managed retrieval
- Git-versioned operational memory today.
- Managed retrieval can be added as customer data and tenant boundaries mature.
Every tool call can become telemetry
- Journal events, tool status, screenshots, errors, and session health already flow through the system.
- Logs, dashboards, and audit trails are first-class scale paths.
Spend maps to measured work
- Model routing can be capped per project, tenant, model, and wave.
- Usage reporting ties agent runs to outcomes, approvals, and verification evidence.
Operating streams
One conductor, two proof lanes: agent execution and revenue assistants.
PhoneAgents is grounded in real, verifiable work. PhoneAgents proves the operating layer handles repo orchestration, PR discipline, and CI visibility. Marketing and sales workflows prove the same conductor turns business context into campaigns, follow-ups, and measurable output. RestaurantBuddy has its own dedicated presentation.
Repo orchestration that can survive real PR work.
PRBranch and worktree disciplineStreams get bounded file ownership, status tracking, review context, and merge-ready handoffs instead of one giant agent thread.
CIChecks stay part of the conversationFailures, logs, screenshots, route smoke tests, and review comments come back to the operator cockpit.
OKApprovals remain explicitPush, merge, deploy, destructive writes, and sensitive config changes stay behind approval gates and audit trails.
LIVEDashboards are operator surfacesThe Mini App shows agents, tasks, skills, approvals, journal events, and health without turning the public website into the app.
Marketing and sales work that compounds into the vault.
ICPContext becomes segmentationVoice notes, customer lists, offers, and channel constraints become briefs, audience maps, and approval-ready outbound plans.
CRMFollow-up is orchestrated workAgents can draft sequences, enrich records, score opportunities, and surface next actions through the same mobile gates.
MKTCampaign output stays measurableLanding page copy, lead magnets, sales assets, and experiments become reusable patterns in the Memory Vault.
SAFEPartner-safe delivery pathCustomer workflows can be reviewed with identity, observability, audit, cost, and reliability discipline.
Mission control and mobile cockpit
Approve, monitor, and ship from your phone.
The cockpit puts agent status, task tracking, approval gates, wave management, skills, and health signals in one mobile-first surface—accessible from any phone, without a second screen.
Launch SheetProject, harness, model, PAM template, and approval mode before dispatch.
Home CockpitAgent dashboard, wave manager, tool status, and approval state in one mobile surface.
Skills MarketplaceReusable agent playbooks with model routing, updates, and direct launch actions.
Public and private surfaces
The website explains the system. The cockpit runs it.
PhoneAgents.com should be a crisp marketing and credibility surface. The operator cockpit remains Telegram plus the Mini App: the place for approvals, status, logs, agent control, screenshots, and recovery decisions.
Marketing, trust, and narrative
- Explains the voice-to-PR operating loop in plain language.
- Shows current phone mockups, PAM flow, ecosystem lanes, and platform architecture and trust positioning.
- Keeps public claims, legal pages, and conversion paths ready for the main launch route while internal previews stay separate.
Operations, approvals, and execution
- Telegram handles voice notes, approvals, alerts, and high-signal status updates.
- The Mini App shows dashboards, agents, tasks, skills, approvals, settings, and journal views.
- Write actions stay policy-bound: request, approve, execute, verify, report.
What happens overnight
Dispatch, monitor, verify, and leave a merge ledger by morning.
The overnight story is concrete: PAM launches streams, watches the live feed, asks for approvals when needed, captures visual QA, and ends with a status report that says what is merged, what is blocked, and what should run next.
20:00
Dispatch the wave
PAM turns the brief into docs, stream prompts, file scopes, approval policy, and verification targets.
22:00
Monitor the agents
The cockpit tracks tool calls, blockers, branch status, context risk, and handoffs without forcing the operator to babysit terminals.
01:00
Run review and visual QA
Code review, smoke checks, screenshots, mobile layout checks, and broken-reference scans become part of the wave.
04:00
Resolve approval gates
Pushes, merges, deploys, and sensitive operations wait for explicit approval while low-risk verification keeps moving.
Morning
Deliver the merge ledger
The report lists merged branches, PRs, changed files, verification results, blockers, and the next suggested streams.
BMad method
Nine phases turn rough intent into verified, shippable waves.
BMad gives PAM a repeatable delivery spine: understand the mission, shape scope, route work, verify the result, and preserve the learning.
Brief
Capture the voice note, documents, screenshots, constraints, and desired outcome.
Clarify
Ask only for missing blockers; infer the rest as explicit assumptions.
Plan
Translate intent into a wave plan, file scopes, acceptance checks, and risk gates.
Dispatch
Launch specialist streams with bounded ownership and no overlapping write scope.
Monitor
Track blockers, context risk, status, tool logs, branch state, and screenshots.
Integrate
Review patches, reconcile handoffs, and keep the operator decision layer intact.
Verify
Run tests, static checks, visual QA, broken-link scans, and review comment passes.
Ship
Prepare commits, PRs, deploy gates, or merge-ready reports with approval discipline.
Remember
Save decisions, failures, and improvements back into Memory Vault and skills.
Not sprints. Waves.
A wave is a time-boxed operating burst with explicit ownership.
PhoneAgents waves are built for parallel AI work: one conductor, clear streams, isolated branches, evidence-based verification, and a report that becomes future memory.
InputsVoice brief, docs, screenshots, repo state, business goal, acceptance criteria.before launch
StreamsImplementation, review, design, research, QA, docs, ops, or partner-prep lanes with separate ownership.parallel
GatesPush, deploy, merge, destructive commands, sensitive config, and production actions require explicit approval.human control
EvidenceTests, screenshots, static scans, PR comments, CI results, changed files, and risk notes.verification
MemoryHandoff report, decisions, reusable prompts, failures, and follow-up actions become vault context.compounding
Onboarding flow
Seven guided steps from first login to first agent wave.
The user should not configure a platform from scratch. PhoneAgents asks for the minimum needed to create a safe first wave, then lets the Memory Vault personalize future defaults.
01Connect Telegram or Mini App.
02Choose project and repo.
03Select harness and model policy.
04Set approval sensitivity.
05Install starter skills.
06Run a guided PAM wave.
07Review report and level up.
Gamification
Gamerscore for operators: wave level-ups, badges, and durable progress.
The points are not decoration. They are a retention surface on top of real operational data: verified waves, memory contributions, shipped PRs, safe approvals, and reusable skills.
First Wave
Launch a PAM-guided stream and complete the handoff.
Ship Night
Finish a wave with tests, screenshots, and a merge-ready report.
Vault Builder
Turn repeated decisions into skills, ADRs, and memory entries.
Operator Rank
Progress reflects real wave maturity, not vanity activity.
Business model
Three revenue lines, one operating layer.
PhoneAgents can monetize the command layer directly while also creating reusable workflows for dev teams and revenue operators.
Operator subscription
Monthly access to mobile cockpit, PAM waves, skills, approvals, memory, and provider routing.
Managed agent execution
Metered agent runs, model usage, review passes, screenshots, and wave capacity with explicit spend controls.
Partner implementation
Setup, workflow design, architecture review prep, skills creation, and business-process automation.
Market
One conductor loop. Multiple markets. One compounding platform.
The market is not only AI coding. The same conductor loop applies anywhere operators need agents, approvals, memory, and measurable work output.
AI agent operations
Teams need orchestration, branch discipline, CI visibility, review handling, and secure approval flows.
Campaign assistants
Briefs, copy, landing pages, lead magnets, segmentation, experiments, and reporting can be coordinated as waves.
Follow-up copilots
Pipeline notes, CRM tasks, outreach sequences, proposal drafts, and customer memory become agent-ready work.
Ecosystem products
One command layer, multiple products, shared intelligence.
PhoneAgents is the operator layer that ties AI development, revenue workflows, agent skills, memory, cloud-ready primitives, and future vertical products into one compounding platform.
AI development cockpit
Mobile control for agent coding waves, PRs, reviews, approvals, and CI visibility.
Second brain
Git-versioned lessons, decisions, skills, and retrieval that reduce cold starts across all products.
Planning and allocation
Turns goals into shippable waves, agent assignments, dependencies, and recovery routes.
Marketing and sales waves
Campaigns, outreach, follow-ups, sales assets, and experiments coordinated through the same command loop.
Operator inbox
Approvals, alerts, status, voice commands, and action buttons stay available on mobile.
Visual command center
Dashboards, agents, tasks, skills, approvals, settings, and wave control in one app shell.
Reusable playbooks
Marketing, coding, research, review, and ops workflows become installable agent capabilities.
Compounding platform
Every additional product writes memory back into the same system and benefits from prior learning. RestaurantBuddy is intentionally handled in its own deck.
Founders preview
Run the brief. Watch the streams. Ship the report.
Be first to run verified AI agent waves from your phone. PAM orchestration, secure cloud-ready infrastructure, and a Memory Vault that makes every wave smarter than the last.