v4.5.7 · Agent runtime today, control plane next

The control plane for autonomous engineering

CodeTether turns objectives into PRDs, coordinates Ralph and specialized swarms, edits real repositories, controls browsers and infrastructure, and keeps every tool call governed by policy.

Start building free View the agent runtime

Rust-native runtimeGitHub App integrationOKR → PRD → RalphSwarm orchestrationMCP + A2AOPA RBAC

live agent run

Objective

Ship GitHub OAuth with policy gates

Objective

Capture the outcome, owner, constraints, and success metric.

PRD

Generate stories, acceptance criteria, dependencies, and quality gates.

Ralph

Iterate in fresh contexts, edit real files, test, and commit.

Control

Stream events, enforce policy, audit tools, and route workers.

$ codetether go "add OAuth login"

creating OKR…

✓ objective: secure social sign-in

✓ PRD: 5 stories, 18 acceptance criteria

✓ swarm: security + docs + tester

↻ ralph: implementing story 4/5

→ policy: OPA authz.rego passed

→ browser: replayed /api/session smoke test

60+

registered tool IDs

∞

workspace context via RLM

release targets

The AI development platform

Autonomous objectives, swarm intelligence, persistent context, multi-modal tools, and production-grade infrastructure — 60+ registered tools, one control plane.

# One command. Objective to implementation.
$ codetether go "Add OAuth2 login with GitHub and Google"

# CodeTether automatically:
# 1. Creates an OKR with measurable outcomes
# 2. Generates a PRD with user stories
# 3. Runs Ralph in isolated implementation loops
# 4. Executes tests, lint, and type checks
# 5. Maps results back to OKR progress
# 6. Returns a ready-to-merge branch

# Plan mode for complex tasks
$ codetether plan "Refactor auth to support SSO"
# Step 1/3: Audit existing auth middleware ✓
# Step 2/3: Add SAML/OIDC providers      ✓
# Step 3/3: Update RBAC policies         ✓
# Ready to merge → feature/sso-refactor

Objective → PRD → Code

Describe an outcome. CodeTether creates an OKR, generates a structured PRD, runs Ralph through each user story, tests changes, and delivers a merge-ready branch.

# One command. Objective to implementation.
$ codetether go "Add OAuth2 login with GitHub and Google"

# CodeTether automatically:
# 1. Creates an OKR with measurable outcomes
# 2. Generates a PRD with user stories
# 3. Runs Ralph in isolated implementation loops
# 4. Executes tests, lint, and type checks
# 5. Maps results back to OKR progress
# 6. Returns a ready-to-merge branch

# Plan mode for complex tasks
$ codetether plan "Refactor auth to support SSO"
# Step 1/3: Audit existing auth middleware ✓
# Step 2/3: Add SAML/OIDC providers      ✓
# Step 3/3: Update RBAC policies         ✓
# Ready to merge → feature/sso-refactor

Swarm & Relay

Spawn specialized sub-agents in parallel or chain them via relay. Each agent has full tool access, its own context, and persistent memory across sessions.

# Parallel swarm execution
const result = await swarmExecute({
  tasks: [
    { name: 'Security', instruction: 'Audit all API routes',
      specialty: 'Security Reviewer' },
    { name: 'Perf', instruction: 'Profile critical path latency',
      specialty: 'Performance Engineer' },
    { name: 'Docs', instruction: 'Generate API reference',
      specialty: 'Technical Writer' }
  ],
  concurrency: 3,
  strategy: 'best_effort'
});

# Relay: chain agents sequentially
relay.delegate({
  target_agent: 'Architect',
  message: 'Design the rate-limiting system'
});
# → Architect hands off to Implementer
# → Implementer hands off to Reviewer
# → Results aggregate automatically

Multi-Modal Tools

Voice, video, browser, podcasts, YouTube, image input, web search, and MCP — 60+ registered tool IDs behind a Rust-native agent runtime.

# Voice: Text-to-speech with cloned voice
await voice.speak({
  text: "Build complete. All tests passing.",
  voice_id: "960f89fc"  // Your cloned voice
});

# Podcast: Generate & publish
const pod = await podcast.create_episode({
  podcast_id: "tech-weekly",
  script: "Today we cover WebAssembly...",
  title: "Ep 42: WASM in Production"
});

# Browser: Headless automation
await browser.goto("https://dashboard.example.com");
await browser.click_text("Deploy");
await browser.wait({ text: "Deployment complete" });
# Capture real app traffic, then replay with patches
await browser.replay({ url_contains: "/api/deploy", body_patch: { env: "prod" } });

# MCP: Connect to any external tool server
const mcpTools = await mcp.listTools({
  command: "npx @modelcontextprotocol/server-postgres"
});

Production Infrastructure

K8s-native deployment with OPA policy engine, RBAC, tenant isolation, and Ed25519 plugin signing. Built in Rust for single-binary performance across 11 platforms.

# Deploy to Kubernetes
$ kubectl apply -f chart/a2a-server/

# OPA Policy Engine — RBAC across your org
policies/
├── authz.rego          # Permission checks
├── api_keys.rego       # Key scope enforcement
├── tenants.rego        # Tenant isolation
└── data.json           # Role→permission mappings

# 11 platform binaries (Linux, macOS, Windows)
$ curl -fsSL https://codetether.run/install.sh | bash
# x64, ARM64, musl, baseline — one binary, zero deps

# Kubernetes tool: manage from the agent
$ kubectl scale --replicas=3 deployment/api
$ kubectl logs --tail=50 pod/worker-7x2a

The Four Pillars

Autonomous AI that codes, orchestrates, and deploys — end to end.

Autonomous Coding

Start with an objective. CodeTether generates the PRD, maps work to stories, runs Ralph, tests, and returns a ready-to-merge branch.

Swarm Orchestration

Spawn focused sub-agents or run best-effort swarms in parallel — security, performance, docs, QA — with shared results aggregated automatically.

Multi-Modal Control

Voice synthesis, podcasts, avatar video, YouTube upload, browser automation with network replay, and MCP bridge to any external tool server.

Enterprise Security

OPA Rego policy engine, RBAC across 5 roles, tenant isolation, Ed25519 plugin signing, and K8s-native deployment.

The AI Development Platform

60+ registered tools, multi-modal AI, persistent context, and enterprise-grade orchestration — all from a single binary.

RLM: Infinite Context
Recursive Language Model processing lets the agent analyze whole repositories, large histories, and sprawling docs with recursive chunking, synthesis, and verification.
Persistent Memory & Recall
Session recall, context browsing, curated memory, and task state keep long-running work coherent across resets, handoffs, and fresh agent iterations.
Swarm Orchestration
Spawn parallel sub-agents with specialized roles. Security review, performance analysis, documentation, QA, and implementation can run concurrently.
Relay Task Routing
Autonomous delegation between agents. Hand off context, chain multi-step workflows, and orchestrate complex pipelines across specialized AI instances.
MCP + Browser Replay
Bridge external MCP servers and control real browsers. Inspect DOM, upload files, capture app traffic, and replay authenticated XHR/fetch requests safely.
GitHub App Integration
Mention @codetether on any issue or PR. Work runs on CodeTether workers — not GitHub Actions runners — and evidence publishes back as Check Runs. Zero runner minutes consumed.
Voice & Media Pipeline
Built-in text-to-speech, transcription, podcast generation, YouTube upload, image input, and AI avatar video. Go from a script to a published asset.

Ralph + RLM Integration

Build Features from a PRD — Automatically

Ralph runs autonomous dev loops: PRD → implement → test → commit → repeat. When context grows too large, RLM compresses it so the loop keeps going.

📄 prd.json

Project

TaskApp

Branch

ralph/task-status

User Stories

○US-001: Add status field to database

○US-002: Display status badge on cards

○US-003: Add status toggle to list

○US-004: Filter tasks by status

RLM Status

Compressions:0

Ralph + RLM Loop

Iterations: 0

Passed: 0/4

Commits: 0

RLM: standby

Click "Start Ralph" to watch autonomous development

Watch how RLM compresses context when it gets too large

Based on Geoffrey Huntley's Ralph pattern

🔄

Autonomous Loop

Ralph implements each story, runs tests, commits code, and moves to the next without stopping.

🗜️

RLM Compression

When memory exceeds threshold, RLM compresses progress while preserving key learnings and file paths.

⚡

Ship Faster, Smarter

Turns PRD into production code without context degradation. This is how you ship without losing quality.

How Ralph + RLM Work Together

📋 PRD Story→💻 Ralph Codes→✓ Tests Pass→📦 Git Commit→📝 Update Memory→🗜️ RLM Compress*→🔁 Next Story

*RLM compression triggers automatically when context exceeds threshold (default: 80K tokens)

Try Ralph + RLM on Your Workspace

Create a PRD → Convert to prd.json → Let Ralph + RLM build it

Unlimited Context. Real Outputs. Reliable Automation.

Most AI degrades as context grows. RLM processes large inputs recursively—breaking work into chunks, verifying results, and producing deliverables.

The Problem: Context Limits

Base models hit context limits on 6–11M token tasks.

RLM completes them with 91.33% accuracy (MIT CSAIL).

The Solution: RLM

RLM treats input as an environment variable, not direct context. The AI writes code to peek, decompose, and recursively call itself on chunks.

RLM(GPT-5) on same tasks:

91.33% accuracy

From failing to solved

How RLM Works

Load as Variable

Your data becomes a variable in a Python REPL, not direct LLM input

Decompose

AI writes code to chunk, filter, and navigate your data programmatically

Recursive Calls

Sub-LLMs process chunks, verify results, and report back

Stitch Output

Results combined into real files: CSV, PDF, code, reports

10M+

tokens processed

100x beyond normal LLMs

91%

accuracy on huge tasks

vs context-limited base models

cheaper than alternatives

median cost lower than base model

RLM is based on peer-reviewed research from MIT CSAIL.

Read the Paper: Recursive Language Models

Zhang, Kraska, Khattab. "Recursive Language Models." MIT CSAIL, 2025.

Background Tasks for Every Workflow

Trigger from dashboard, API, or automation platform. Kick back. Get real deliverables in your inbox.

🔄

Engineering Teams

Write a PRD, run Ralph. Wake up to implemented features with tests passing and commits pushed. RLM ensures infinite context so your full workspace is understood, not just single files.

PRD → Ralph implements → Review PR in morning

Autonomous dev
PRD → Code
RLM context

🎓

Course Creators

Stop spending hours on landing pages. Tell the agent "build me a sales page for my new course" and get production-ready HTML in minutes. It reads your existing content, matches your brand, and writes converting copy.

New course idea → Agent builds funnel → You review & launch

Landing pages
Sales copy
Email sequences

🏢

Agency Owners

Your clients need reports, audits, and content. RLM agents analyze entire websites, generate SEO reports, and write blog posts—all triggered from your project management tool via webhook.

Client request → n8n triggers agent → Deliverable ready

Client reports
SEO audits
Content at scale

🛒

E-commerce Sellers

Product descriptions that convert. Feed the agent your inventory spreadsheet and get optimized listings for Amazon, Shopify, and Etsy. Bulk processing hundreds of SKUs overnight.

CSV upload → Agent writes listings → Export to platforms

Product descriptions
Listing optimization
Bulk processing

⚡

No-Code Automators

18 Zapier components connect CodeTether to 6,000+ apps. Trigger autonomous tasks from Airtable, Slack, or Gmail. No code required—just connect and configure.

Zap trigger → Agent runs task → Result to destination

Zapier triggers
18 components
No-code setup

⚙️

DevOps & Infrastructure

Agents that monitor your systems, manage deployments, and self-heal on Kubernetes. Ed25519-signed plugins ensure only trusted tools touch production. Every action audit-logged for compliance.

Alert fires → Agent diagnoses → Auto-remediation + audit log

K8s management
Incident response
Audit compliance

💻

SaaS Founders

Your product workspace is growing. RLM agents can analyze your full source context, find bugs, write tests, and generate documentation across teams and tools.

Push to GitHub → Agent reviews → Issues created

Code review
Test generation
Documentation

✍️

Content Marketers

Research, outline, write, optimize. Agents that read your competitors' content, analyze what ranks, and produce SEO-optimized articles. Triggered from your editorial calendar.

Calendar event → Agent researches & writes → Ready for review

Blog posts
Research reports
Social content

🎯

Consultants & Coaches

Client intake forms trigger personalized analysis. The agent reads their questionnaire, analyzes their situation, and drafts a custom proposal—before your first call.

Form submission → Agent analyzes → Proposal in inbox

Proposal generation
Client analysis
Custom reports

The Road to the Agent Control Plane

The agent runtime is already deep. Next is making it governable, observable, and programmable for teams.

ShippedLive

Core Platform

60+ registered tools, autonomous coding loops, multi-agent orchestration, and multi-modal output — live in the agent runtime.

OKR tracking and execution runs
OKR → PRD → Ralph autonomous pipeline
Structured PRD generation and validation
Ralph iterative implementation loop
RLM infinite context processing
RLM answer verification
Session recall and context browsing
Persistent memory across sessions
Real file output (CSV, PDF, code)
Plan mode, safe edit previews, and undo
Browser control with network replay
Web search, web fetch, and code search
Voice TTS and transcription
Podcast generation & RSS
YouTube upload
AI avatar video generation
Image input for vision-capable models
MCP bridge (Model Context Protocol)
Kubernetes pod management
Sub-agent relay autochat
Swarm parallel execution
Multi-model orchestration
OPA Rego RBAC and tenant isolation
Self-hosted (Docker / K8s / bare metal)

Control Plane API

Expose the full agent runtime as a durable, observable, multi-tenant control plane.

Unified control plane API for OKRs, PRDs, runs, tools, and sessions
Run lifecycle events and streaming progress
Dashboard for OKR/run monitoring and replayable audit trails
Team workspaces with role-scoped agent permissions
Scheduled and webhook-triggered autonomous runs
Worker pools for local, cloud, and Kubernetes executors
Tool catalog, policy gates, and per-tenant capability controls
Pre-built workflow templates for engineering, ops, and media
Bulk task API and batched background execution
Zapier, n8n, Make, and GitHub app integrations
Usage metering, quotas, and cost controls

Future

Autonomous Organization Layer

Turn goals, policies, tools, and workers into a self-improving execution fabric.

Policy-aware autonomous delegation across teams and tenants
Learning workflow library from successful runs
Organization-level OKR decomposition into agent run plans
Cross-repository dependency reasoning and migration planning
RLM deeper recursion and domain-trained retrieval models
Custom fine-tuned coding and operations models
Real-time collaborative voice agent
Video understanding, generation, and review loops
Marketplace for signed agent skills and MCP tool packs
White-label and private-cloud deployments for agencies and enterprises

Want a feature? We listen to our users.

Request a feature

Simple Pricing. Real Results.

Start free. Upgrade when you're hooked. Cancel anytime.

Free

Get started free. Perfect for testing workflows.

Best for: Testing & experimentation

$5 token credit included

100

tasks/mo

workers

3 workspaces

workspaces

100 tasks / month
1 agent
3 workspaces
$5 prepaid token credit
Per-token usage tracking
HMAC-SHA256 auth + full audit trail
Community Discord

Start Free

ProMost Popular

$49/mo

For builders who need serious throughput.

Best for: Solopreneurs & automation builders

$50 token credit / month included

5,000

tasks/mo

workers

20 workspaces

workspaces

5,000 tasks / month
5 concurrent agents
20 workspaces
$50/mo prepaid token credit
Per-model cost breakdown
Monthly spending limits
OPA Rego RBAC policies
Ed25519 plugin signing
RLM unlimited context
Swarm orchestration
MCP tool bridge

Enterprise

$199/mo

Unlimited tasks, workers, and workspaces.

Best for: Teams & agencies

$150 token credit / month included

∞

tasks/mo

∞

workers

Unlimited

workspaces

Unlimited tasks / month
Unlimited agents
Unlimited workspaces
$150/mo prepaid token credit
Per-tenant token isolation
Custom spending limits & alerts
Everything in Pro
K8s self-deployment
Team workspaces with RBAC
Relay multi-agent pipelines
Voice & media pipeline
Custom OPA security policies
Dedicated support

Transparent Token-Based Billing

Every plan includes prepaid token credits. You only pay for what your agents actually use, tracked per-request with full model-level cost breakdowns.

Per-Token

Billed by actual input, output, cache & reasoning tokens

Per-Model

Each AI model priced at cost. No hidden markup.

Prepaid

Credits included with your plan. Add more anytime.

Limits

Set monthly spending caps to stay in budget.

Model Pricing (per 1M tokens)

Model	Input	Output	Cache Read
Claude Opus 4.6	$15.00	$75.00	$1.50
Claude Opus 4.5	$5.00	$25.00	$0.50
Claude Sonnet 4	$3.00	$15.00	$0.30
GPT-5.2	$3.00	$12.00	—
GPT-4.1	$2.00	$8.00	—
Gemini 3 Pro	$1.50	$12.00	—
Gemini 2.5 Flash	$0.15	$0.60	—

How It Works

1. Describe

CLI, API, or Dashboard

Describe what you want built in plain English

2. Orchestrate

Autonomous AI Agents

Ralph generates PRDs, spawns swarms, and implements stories

3. Ship

Ready-to-Merge Code

Auto-committed, auto-tested PR waiting in your branch

14-day money-back guarantee. No questions asked.

Build with the agent runtime now

Start with one objective. End with reviewed, tested work.

Use CodeTether locally, self-host it, or connect it to the cloud control plane as it rolls out. The runtime is already capable; the API surface is becoming the system of record.

Start building free View on GitHub

What you get

Generate PRDs from rough objectives
Run Ralph loops against real repos
Spawn focused reviewer/tester/doc swarms
Govern tools with OPA and audit trails

Frequently Asked Questions

- How is CodeTether different from OpenClaw?
  OpenClaw is a Node.js gateway that shipped with auth: none by default. CodeTether is a Rust-based perpetual cognition runtime with mandatory HMAC-SHA256 authentication, Ed25519 code-signed sandboxed plugins, persona swarms with scoped permissions, and append-only audit logging. It self-deploys on Kubernetes. These are shipped features in v1.1.0, not a roadmap.
- Why is CodeTether written in Rust?
  An autonomous agent with system-level access IS infrastructure. Memory safety isn't optional when your agent can read files, execute commands, and take actions on its own. You wouldn't write a database in JavaScript — you shouldn't write an autonomous agent in it either.
- What are Persona Swarms?
  Instead of one bot, you get a coordinated team of agents — monitoring, deployment, code review — each with its own scoped permissions and security boundary. A compromised monitor can't escalate to deploy access. That's not paranoia, that's basic engineering.
- What is perpetual cognition?
  Most AI agents are request-response: you ask, they answer. CodeTether runs continuous thought loops that persist across restarts, survive node failures, and scale horizontally. Your agents reason, plan, and act autonomously — not just when you type something.
- What is Open Policy Agent (OPA)?
  OPA is a CNCF-graduated policy engine that CodeTether uses for fine-grained authorization. We ship Rego policies for API authorization, tenant isolation, and API key scoping — all declarative, testable, and auditable. You can write custom Rego rules to control exactly what each agent, user, or API key is allowed to do.
- Can agents modify their own code?
  Yes. CodeTether agents can write code, open PRs, and evolve their capabilities. But every modification goes through an auditable gate you define. Every decision is logged, every action is traceable. The agents get smarter while you stay in control.
- What is RLM?
  RLM (Recursive Language Models) is MIT research that treats input as an environment variable, not direct LLM context. Instead of stuffing everything into one prompt, RLM recursively decomposes tasks, processes chunks independently, verifies results, and stitches them together. CodeTether uses RLM to handle 10M+ tokens.
- Can I self-host?
  Yes. CodeTether is open source under the MIT License. Run it on your own servers with full control. Docker images and Helm charts provided. Your infrastructure shouldn't depend on someone else's business model.
- How does the security model work?
  Authentication uses HMAC-SHA256 bearer tokens and is mandatory — there is no flag to disable it. All plugins are sandboxed with Ed25519 code signing and SHA-256 integrity checks. Every action is logged to an append-only JSON Lines audit trail with timestamps, categories, and metadata. Persona swarms enforce least-privilege — each agent only has the permissions it needs. Enterprise plans can add custom OPA Rego policies.
- Is there a managed/hosted option?
  Yes. Self-host for free, or use our managed platform with tiered pricing: Free (100 tasks/mo + $5 token credit), Pro ($49/mo, 5,000 tasks + $50/mo token credit), and Enterprise ($199/mo, unlimited tasks + $150/mo token credit). All plans include the same security guarantees.
- How does token billing work?
  Every AI request consumes tokens — input tokens (your prompt), output tokens (the response), and sometimes cache or reasoning tokens. We track usage per-request at the model level and deduct from your prepaid credit balance. You can see exactly what each model costs per million tokens, set monthly spending limits, and add credits any time. No surprises.

Ready to Deploy Agents You Can Actually Trust?

Tell us about your infrastructure. We'll help you deploy CodeTether.

The control plane for autonomous engineering

Ship GitHub OAuth with policy gates

The AI development platform

Objective → PRD → Code

Swarm & Relay

Multi-Modal Tools

Production Infrastructure

Objective → PRD → Code

Swarm & Relay

Multi-Modal Tools

Production Infrastructure

The Four Pillars

Autonomous Coding

Swarm Orchestration

Multi-Modal Control

Enterprise Security

The AI Development Platform

RLM: Infinite Context

Persistent Memory & Recall

Swarm Orchestration

Relay Task Routing

MCP + Browser Replay

GitHub App Integration

Voice & Media Pipeline

Build Features from a PRD — Automatically

📄 prd.json

Autonomous Loop

RLM Compression

Ship Faster, Smarter

How Ralph + RLM Work Together

Unlimited Context. Real Outputs. Reliable Automation.

The Problem: Context Limits

The Solution: RLM

How RLM Works

Load as Variable

Decompose

Recursive Calls

Stitch Output

Background Tasks for Every Workflow

Engineering Teams

Course Creators

Agency Owners

E-commerce Sellers

No-Code Automators

DevOps & Infrastructure

SaaS Founders

Content Marketers

Consultants & Coaches

The Road to the Agent Control Plane

Core Platform

Control Plane API

Autonomous Organization Layer

Simple Pricing. Real Results.

Free

ProMost Popular

Enterprise

Transparent Token-Based Billing

Model Pricing (per 1M tokens)

How It Works

Start with one objective. End with reviewed, tested work.

Frequently Asked Questions

How is CodeTether different from OpenClaw?

Why is CodeTether written in Rust?

What are Persona Swarms?

What is perpetual cognition?

What is Open Policy Agent (OPA)?

Can agents modify their own code?

What is RLM?

Can I self-host?

How does the security model work?

Is there a managed/hosted option?

How does token billing work?

Ready to Deploy Agents You Can Actually Trust?

The AI development platform

Objective → PRD → Code

Swarm & Relay

Multi-Modal Tools

Production Infrastructure

Objective → PRD → Code

Swarm & Relay