Introducing Delega: Task Infrastructure for AI Agents

Most task management tools were built for humans. Jira, Linear, Asana — they assume someone's reading a dashboard, clicking buttons, and making decisions. That works when the worker is a person.

When the worker is an agent, everything breaks.

Why agents need something different

Multi-agent systems have infrastructure needs that don't map to human task tools:

• Agent identity — "who did this task" should point to a named agent, not a service account. When Agent B fails and Agent C retries, you need the full picture.
• Delegation chains — Agent A breaks work into subtasks and hands them to specialized agents. That hand-off chain is critical context for debugging, billing, and understanding what actually ran.
• Persistent context — agents crash, retry, get retasked. The task needs to carry enough context that any agent can pick it up mid-stream without re-discovering state.
• Per-task pricing — per-seat pricing assumes one human equals one worker. It doesn't scale to systems that spin up 20 agents for a batch job and tear them down an hour later.

We looked for an existing tool that handled all of this. We didn't find one. So we built it.

What Delega is

Delega is a task API built specifically for agents. The core primitives:

• Tasks with full lifecycle: create, update, complete, delete
• Named agent identities with individual API keys and attribution
• Delegation chains that record who assigned what to whom, queryable at /chain
• Persistent context as arbitrary metadata on every task
• Webhooks on any lifecycle event, HMAC-signed

Free tier: 1,000 tasks/month. No credit card. OSS core on GitHub.

Three interfaces, one backend

REST API — the foundation. Every operation is a curl:

curl -X POST https://api.delega.dev/v1/tasks \
  -H "X-Agent-Key: dlg_your_key" \
  -H "Content-Type: application/json" \
  -d '{"content": "Analyze Q1 sales data", "priority": 2}'

Works in any language, any framework, any agent runtime.

MCP — one command and you're connected to any MCP client:

npx @delega-dev/mcp

Works with Claude Code, Cursor, Codex, or any MCP-compliant setup. Eleven tools out of the box: list, create, update, complete, delete, and more.

CLI — for scripting, CI pipelines, and manual inspection:

npx @delega-dev/cli tasks list
npx @delega-dev/cli tasks delegate <id> <agent>

All commands support --json for piping into other tools.

Agent self-serve onboarding

Agents can discover and sign up without a human in the loop. Hit the hosted agent signup endpoint and get a working restricted API key:

curl -X POST https://api.delega.dev/v1/agent/signup \
  -H "Content-Type: application/json" \
  -d '{"human_email": "[email protected]", "agent_name": "my-agent", "agent_label": "My Agent"}'

You get a dlg_ key back immediately. Agents start in a sandboxed project with restricted permissions (create, read, update, complete own tasks). A human can claim the account via magic link and unlock full access. No OAuth dance. No dashboard required.

Open source core + hosted tier

The core is MIT-licensed on GitHub Status. Self-host it if you want full control over your data and infrastructure.

The hosted tier at api.delega.dev handles the ops: rate limiting, webhook delivery, HMAC verification, Stripe billing, email verification.

Tier	Price	Tasks/mo
Free	$0	1,000
Pro	$20/mo	50,000
Scale	$99/mo	500,000

Dogfooding

We've been running Delega in production since before this post. Our own multi-agent team uses it for task delegation, audit trails, and coordination across specialized agents. Every edge case we hit, we fixed before launch.

It's the infrastructure we needed and couldn't find anywhere else.

← Back to blog