agentctl — Secure & lightweight agents, from edge to cloud

[agentctl]

0:home* 1:edge 2:cloud 3:logs

| ● COMING SOON | 00:00:00

HEROPane 0 — bash

agentctl@home $ agentctl --about

┌──────────────────────────────────────────────┐ │ │ │ ▄▀█ █▀▀ █▀▀ █▄░█ ▀█▀ █▀▀ ▀█▀ █░░ │ │ █▀█ █▄█ ██▄ █░▀█ ░█░ █▄▄ ░█░ █▄▄ │ │ │ │ Define agents. Ship anywhere. │ └──────────────────────────────────────────────┘

"Secure & lightweight agents, from edge to cloud."

● COMING SOON

curl -sSL https://agentctl.sh/install | sh

~10MBBinary Size

2sCold Start

5+Providers

0Dependencies

[INFO] agent "sre-bot" ready — claude-sonnet-4-20250514 via anthropic

[INFO] agent "edge-vision" ready — llama3.2-vision via llama.cpp

[INFO] daemon "file-watcher" started — watching /var/uploads

[INFO] egress gateway active — 3 rules loaded

[INFO] workflow "security-scan" pipeline — 3 stages ready

AGENT REGISTRYPane 1 — bash

agentctl@home $agentctl status

Agent	Model	Provider	Runtime	Tools	Memory	Status
sre-bot	claude-sonnet-4-20250514	anthropic	cloud/ecs	12	redis	running
code-reviewer	gpt-4o	openai	cloud/ecs	8	sqlite	running
edge-vision	llama3.2-vision	llama.cpp	edge/rpi5	4	in-memory	running
summarizer	mistral-7b	vllm	cloud/gpu	3	redis	idle
security-scan	claude-haiku	anthropic	edge/laptop	6	sqlite	stopped

DEPLOYPane 2 — bash

$agentctl deploy aws

Deploying support-agent to us-east-1 ...

✓

Parsing agent.toml

3 tools, 2 providers, memory: redis

✓

Building container image

agentctl:v0.8.2-slim (12MB)

✓

Pushing to ECR

123456789.dkr.ecr.us-east-1.amazonaws.com

✓

Creating CloudFormation stack

agentctl-support-agent-prod CREATE_COMPLETE

✓

ECS task registered and running

Fargate 0.25 vCPU / 512MB, desired: 1

✓

ALB health check passing

200 OK — /healthz (p99: 12ms)

✓

CloudWatch log group configured

/agentctl/support-agent/prod

Endpoint:https://support-agent.agentctl.example.com

API key:ak_live_••••••••••••d7f3

Est. cost:~$40/month (Fargate + ALB)

✓ Deploy complete in 47s

ARCHITECTURE Pane 3 — agentctl architecture

$ agentctl architecture

PCB-style system diagram

One config. Two runtimes. Your infrastructure.

CONFIGPane 4

$cat agent.toml

# that's your entire agent [agent] name = "incident-responder" description = "Diagnose production issues" [model] provider = "anthropic" model = "claude-sonnet-4-20250514" max_iterations = 10 [[tools]] name = "prod_db" type = "database" read_only = true [[tools]] name = "pagerduty" type = "http" base_url = "https://api.pagerduty.com" [[tools]] name = "shell" type = "shell" allowed_commands = ["kubectl", "docker"] timeout_seconds = 30 [memory] backend = "redis" ttl = "24h" # long_term = "mem0" # coming soon — semantic recall across sessions [security] egress_allowlist = ["api.pagerduty.com", "db:5432"] block_internal_networks = true

CHATPane 5

$agentctl chat sre-bot

user:

"API latency spiked on /v2/orders — what's going on?"

sre-bot:

thinking...

tool:prod_db — SELECT avg(latency_ms) FROM requests WHERE endpoint = '/v2/orders' AND ts > now() - interval '1h'

tool:pagerduty — GET /incidents?status=triggered

sre-bot:

"Found it. The orders DB connection pool hit its limit 47 min ago. Average latency jumped from 120ms to 2.4s. I see a related PagerDuty incident PD-4891. Recommend scaling the pool from 20 to 50 connections."

sre-bot>

WORKFLOW PIPELINEPane 6 — bash

$agentctl workflow --show security-cam

Pipeline: security-cam — 3 stages — edge+cloud hybrid

motion-detect

llama3.2-vision

edge / rpi5

──▶

vision-analyze

claude-sonnet-4-20250514

cloud / ecs

──▶

responder

llama3.1-8b

edge / laptop

Last run: 2m ago · Avg latency: 3.2s · Runs today: 847

DAEMON MODEPane 7

$agentctl daemon list

sre-botwebhookPOST /hooks/sre up 14d 6h

edge-visionfile-watch/var/camera/frames up 3d 11h

summarizercron0 9 * * MON-FRI● idle · next 9am

SECURITY LEDGERPane 8

$agentctl ledger --tail 5

ALLOWEDsre-bot→ api.pagerduty.com GET /incidents2s ago

BLOCKEDsre-bot→ pastebin.com POST (egress denied)14s ago

ALLOWEDsre-bot→ prod_db SELECT avg(latency_ms)...18s ago

BLOCKEDcode-rev→ leaked API key detected (scrubbed)3m ago

ALLOWEDedge-vis→ vision-analyze (pipeline handoff)4m ago

LIVE LOGSPane 9 — $ agentctl logs -f

FEATURESPane 10 — bash

$agentctl features

AGENTCTL FEATURES — what's in the box

reactReAct reasoning loop — observe, think, act, repeat

workflowMulti-agent pipelines — chain agents across cloud & edge

daemonBackground mode — cron, webhooks, file-watch triggers

sandbox-runtimeIsolated execution — seccomp, namespace, resource limits

egressEgress gateway — whitelist outbound domains per agent

ledgerRequest ledger — every API call logged & auditable

edgeLocal inference — llama.cpp, Metal/CUDA, fully offline

visualizerEmbedded web UI — inspect agent state, memory, tool calls

deployOne-command deploy — ECS, Raspberry Pi, or bare metal

memoryPluggable memory — Redis, SQLite, in-memory backends

long-term-memCustom semantic memory engine — entities, summaries, recall across sessions

built-in toolsVision, web search, code execution — zero config, works out of the box

PROVIDERSPane 11 — bash

$agentctl providers --list

Provider	Type	Models	Latency	Status
Anthropic	cloud	claude-sonnet, haiku, opus	180ms	ok
OpenAI	cloud	gpt-4o, gpt-4o-mini, o1	210ms	ok
vLLM	cloud	mistral, llama, custom	95ms	ok
llama.cpp	local	llama3, phi, mistral, vision	~800ms	ok

Tip: Use provider = "llama.cpp" for air-gapped edge deployments. No API key needed.

WAITLISTEarly Access

Get Early Access

Join the waitlist. Be first to deploy agents from a single binary.

"agentctl" — 12 panes | GitHub | Docs | Research

rust | open-source | 00:00:00