HomeToolkitGovernanceServicesPricing
The toolkit

Everything your engineers and their agents need for AI, in one toolkit.

Skills, harnesses, models, agent building and tracing, MCP access and guardrails that work inside Claude Code, Codex, Cursor and Copilot. One install, one contract.

Your team’s toolkit01 / Connect
$ npx codiedev@latest connect
Starter packOne toolkit
Approved modelsOne toolkit
MCP connectorsOne toolkit
Guardrail hooksOne toolkit
Your company’s rules, in the tools you use.
Your tools
Claude CodeCursorCodexGitHub CopilotOpenCodeYour agentAPI
Your models
ClaudeCodexGeminiKimiGLMMeta Muse
01 / Skills

Skills your whole team installs on day one, with proof of which ones ship.

Your best engineers have already figured out how to get the most from AI. Codiedev turns their skills, hooks and agents into starter packs anyone can install with one command, keeps them versioned, and shows which ones actually lead to merged code.

Starter pack

New engineer onboarding

Five skills. Hooks. Agents.

  • Write a ticket
  • Plan the change
  • Review a PR
  • Write tests
  • QA checklist
StableBeta
01Experimental

Try it on real work

02Team standard

Promote what works

03Company standard

Share it company-wide

Usage and merged PRs guide each promotion.

  • Starter packs. Bundle skills, hooks and agents for a role or team: new-engineer onboarding, ticket writing, PR review, test writing, QA. One install into Claude Code, Codex, Cursor, Copilot or Codie, with stable and beta channels.
  • Versioned registry. Every push is versioned. Teams that fork a skill see when the original moves on.
  • Curated marketplace. Vetted skills from reputable authors, cached in your own registry. Fork popular skills from other teams in one click.
  • Governance ladder. Promote skills from experimental to team standard to company standard, based on evidence instead of opinion.
  • Proof they ship. See which skills ran in the sessions that became merged PRs, including review and test skills on each PR.
02 / Harnesses

Keep Claude Code, Codex, Cursor or Copilot. Or use Codie to reach any approved model.

Nothing changes for the engineer. Codiedev adds skills, capture and guardrails to the tools you already pay for, and fills the gaps those tools leave.

Your current tools

Skills, capture and guardrails

Claude Code · Codex · Cursor · Copilot

Codie

Our managed harness

Your approved models and prices
New approved models, no setup.

Any harness

The same approved catalog

pi · OpenCode · Crush · Grok CLI

The tools you already pay for. Claude Code, Codex, Cursor and Copilot. Codiedev adds skills, session capture and security hooks behind the scenes.

Codie, our managed harness. Built on OpenCode. Every launch pulls your company's approved models and prices, so new models appear with no setup. Only have an Anthropic deal? Your engineers still get OpenAI and open models in Codie.

Any harness. pi, OpenCode, Crush and Grok CLI, locked to your approved models, with analytics to track adoption.

03 / Models and policy

Every new AI model, available the day it clears your security policy.

Set your rules once. We test new models daily against them, and admins get a weekly digest of the ones ready to switch on. Engineers only ever see what's approved.

01

Your policy

Retention, training, origin and regions

02

We source

New models tested daily

03

Admins toggle

By team, role or person

04

Engineers pick

Only what is approved

  1. Your policy. Zero data retention, no training on your data, excluded countries of origin, and US-only or US + EU regions.
  2. We source. New models tested daily against your policy.
  3. Admins toggle. Turn models on or off by team, role or person, with budgets attached.
  4. Engineers pick. From the approved list, never beyond it.
Your BedrockVia Codiedev

Configure it model by model.

  • Failover. Requests reroute across approved providers on rate limits and outages, with daily route health.
  • Lifecycle. When a model retires, engineers get a suggested replacement and switch with one click.
  • Your keys, ours, or both. Connect your own AWS Bedrock, Vertex or Azure accounts and keep burning down those commits, then fill the gaps with models from Codiedev's vendor contracts and keys. Configure it model by model. No token markup either way.
04 / Agent building and tracing

Build agents and trace every run, on the same platform.

Your teams are building agents for code review, triage and more. Point them at Codiedev and the same policies, budgets and failover apply to them as to your engineers, with a full trace of every run.

One API
from openai import OpenAI

client = OpenAI(
    base_url="https://gateway.codiedev.com/api/v1",
    api_key=service_key,
)
Run timeline

Every step, in view.

WorkloadEnvironmentPR
Request received
Approved model selected
Tools called
Result and cost recorded

Build agents

  • OpenAI-compatible LLM API. Swap the base URL in any OpenAI SDK. Streaming and tool calls included.
  • Direct to inference providers. Open models from Fireworks, Baseten and Together; closed models through Bedrock, Vertex and Azure.
  • Service keys for agents. Per-key model allowlist, monthly budget, rate limit, expiry and rotation.

Trace every run

  • Spend per agent and run. Cost, tokens and errors for every run, by workload and environment.
  • Run timelines. Step-by-step runs, prompts and failure detection.
  • Cheaper-model suggestions. See when another approved model would do the same work for less.
05 / MCP gateway Early access

One MCP gateway for Jira, Salesforce and more, with access and spend limits by team.

Every developer's agents reach your business tools through one gateway, so you decide who can connect to what and how much they can spend.

Your agents

Engineers & teams

codiedev MCP gateway

Team access · Monthly limits

Warn at 90%. Block oversized calls.
Jira
Salesforce
And more
  • Team access. Turn each connector on per team. Everyone else can request access.
  • Spend guards. Monthly limits per person, a warning at 90%, and oversized calls blocked.
  • Cost visibility. Cost per active person and your heaviest users, with call sizes logged for audit.
06 / Guardrails

Guardrails that block secrets and destructive commands in every harness.

The same security hooks run in Claude Code, Codex, Cursor, Copilot and Codie, so a rule you set once applies everywhere your engineers work.

  • Secrets stay out of prompts. Keys and credentials are caught before they leave the machine.
  • Destructive commands blocked. Risky shell and git commands stopped before they run.
  • One rule set, every tool. No per-harness configuration to keep in sync.
Guardrails / Every harness
agent → git push --force origin main
Blocked before execution

Force-pushing the protected branch violates your company policy.

Secrets stay out of prompts. Rules stay in sync.
What each engineer sees

Engineers see their own numbers, and a cheaper model that does the same work.

Every engineer gets a personal view of their sessions, limits, token use and merged work, plus what their teammates' most effective skills and models are. It's a coaching tool, not a scoreboard.

Developer model usage showing personal costs, tokens and model usage shares. Demonstration data.

Give one team the full toolkit.

We'll set up starter packs, approved models and guardrails for a pilot team, and show you what ships.

See Governance
Workshops and workflow consulting →