inngest-agents

Name: inngest-agents
Author: inngest

134installs5forks

How to install inngest-agents?

inngest/inngest-skills/inngest-agents

$npx -y skills add inngest/inngest-skills --skill inngest-agents

Installs into the current project.

Summary

Use when building durable AI agents or agentic workflows with Inngest and AgentKit, including model calls, tool calls, multi-agent networks, human approval, realtime progress, provider rate limits, crash-safe execution, and Agent Evals handoff. Covers AgentKit, step.ai, `step.r

Files · 1

View on GitHub

SKILL.md

1	# Inngest Agents
2
3	Use this skill when the user wants to build, migrate, or debug an AI agent,
4	multi-step AI workflow, tool-calling loop, support agent, research agent,
5	human-in-the-loop review flow, or realtime agent UI.
6
7	Inngest's AgentKit defines agents with `createAgent`; when an AgentKit run is
8	owned by an Inngest function, model calls use Inngest `step.ai` so they retry
9	and cache model results durably. Use the lower-level Inngest step primitives
10	around the agent for database reads/writes, tool side effects, waits,
11	approvals, realtime progress, and flow control.
12
13	Official references:
14
15	- AgentKit agents: https://agentkit.inngest.com/concepts/agents
16	- `createAgent`: https://agentkit.inngest.com/reference/create-agent
17	- AI inference and `step.ai`: https://www.inngest.com/docs/features/inngest-functions/steps-workflows/step-ai-orchestration
18	- Agent Evals: https://www.inngest.com/docs/learn/agent-evals
19	- AgentKit realtime hooks: https://www.inngest.com/changelog/2025-09-24-agentkit-use-agent
20
21	## Copyable Example
22
23	When starting a durable support or tool-calling agent from scratch, inspect the
24	companion example at `../../examples/durable-agent`. It shows the expected
25	agent-first shape: quick HTTP trigger, typed events, AgentKit inside an
26	Inngest function, step-scoped context loading, human approval with
27	`step.waitForEvent`, and durable side effects after approval.
28
29	## When to Use Inngest for Agents
30
31	Good fit:
32
33	- Agent can take longer than one HTTP request.
34	- Agent calls tools, APIs, databases, browsers, sandboxes, or MCP servers.
35	- Agent needs to survive deploys, crashes, serverless timeouts, or model/API
36	failures.
37	- Agent may wait for human approval, external callbacks, scheduled follow-up,
38	or user input.
39	- Agent progress should stream to a UI from the durable workflow.
40	- Model/provider calls need concurrency or throttle limits.
41	- Duplicate sends, charges, writes, or model calls would be costly.
42
43	Not usually worth it:
44
45	- One short, read-only model call with no side effects and no need for durable
46	progress.
47	- UI-only autocomplete where losing the request is acceptable.
48
49	## Architecture
50
51	Use this shape unless the repo already has a stronger established pattern:
52
53	1. The HTTP/server action layer validates auth, stores the user's intent if
54	needed, emits an event with a stable `id`, and returns quickly.
55	2. An Inngest function owns the agent run.
56	3. Load state and external context inside `step.run`.
57	4. Create AgentKit agents inside the function or import agent/network
58	factories.
59	5. Run model inference through AgentKit / `step.ai`; wrap non-model tool side
60	effects in `step.run`.
61	6. Use `step.waitForEvent` or `step.waitForSignal` for human approval and
62	external callbacks.
63	7. Publish durable progress with native realtime.
64	8. Add sessions and scores when the agent outcome needs to be evaluated later.
65	9. Apply flow control at the function level for provider and tenant limits.
66
67	## Basic AgentKit Function
68
69	Prefer a small, typed function first; add networks and extra tools after the
70	single-agent path is proven.
71
72	```typescript
73	import { createAgent, openai } from "@inngest/agent-kit";
74	import { inngest } from "@/inngest/client";
75
76	export const summarizeTicket = inngest.createFunction(
77	{
78	id: "summarize-ticket",
79	triggers: [{ event: "support/ticket.created" }],
80	concurrency: [{ key: "event.data.accountId", limit: 2 }]
81	},
82	async ({ event, step }) => {
83	const ticket = await step.run("load-ticket", () => {
84	return getTicket(event.data.ticketId);
85	});
86
87	const writer = createAgent({
88	name: "support-summary-writer",
89	system: "Write a concise support-ticket summary with next actions.",
90	model: openai({ model: "gpt-4o" })
91	});
92
93	const { output } = await writer.run(JSON.stringify(ticket));
94
95	await step.run("save-summary", () => {
96	return saveTicketSummary(event.data.ticketId, output);
97	});
98
99	return { ticketId: event.data.ticketId };
100	}
101	);
102	```
103
104	## Tool Calls
105
106	Tools can be defined with AgentKit, but agent-safe tools should still follow
107	durability rules:
108
109	- Read-only tool calls can run as part of the agent when replaying is harmless.
110	- External side effects should be isolated with stable IDs and `step.run`
111	boundaries, or implemented as tool handlers that use the provided `step`.
112	- Tool outputs should be small enough for step state limits.
113	- Validate tool parameters

People who liked this also viewed

6 picks

Type

inngest-agents

134installs5forks

View on GitHub

How to install inngest-agents?

inngest/inngest-skills/inngest-agents

$npx -y skills add inngest/inngest-skills --skill inngest-agents

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Inngest Agents
2
3	Use this skill when the user wants to build, migrate, or debug an AI agent,
4	multi-step AI workflow, tool-calling loop, support agent, research agent,
5	human-in-the-loop review flow, or realtime agent UI.
6
7	Inngest's AgentKit defines agents with `createAgent`; when an AgentKit run is
8	owned by an Inngest function, model calls use Inngest `step.ai` so they retry
9	and cache model results durably. Use the lower-level Inngest step primitives
10	around the agent for database reads/writes, tool side effects, waits,
11	approvals, realtime progress, and flow control.
12
13	Official references:
14
15	- AgentKit agents: https://agentkit.inngest.com/concepts/agents
16	- `createAgent`: https://agentkit.inngest.com/reference/create-agent
17	- AI inference and `step.ai`: https://www.inngest.com/docs/features/inngest-functions/steps-workflows/step-ai-orchestration
18	- Agent Evals: https://www.inngest.com/docs/learn/agent-evals
19	- AgentKit realtime hooks: https://www.inngest.com/changelog/2025-09-24-agentkit-use-agent
20
21	## Copyable Example
22
23	When starting a durable support or tool-calling agent from scratch, inspect the
24	companion example at `../../examples/durable-agent`. It shows the expected
25	agent-first shape: quick HTTP trigger, typed events, AgentKit inside an
26	Inngest function, step-scoped context loading, human approval with
27	`step.waitForEvent`, and durable side effects after approval.
28
29	## When to Use Inngest for Agents
30
31	Good fit:
32
33	- Agent can take longer than one HTTP request.
34	- Agent calls tools, APIs, databases, browsers, sandboxes, or MCP servers.
35	- Agent needs to survive deploys, crashes, serverless timeouts, or model/API
36	failures.
37	- Agent may wait for human approval, external callbacks, scheduled follow-up,
38	or user input.
39	- Agent progress should stream to a UI from the durable workflow.
40	- Model/provider calls need concurrency or throttle limits.
41	- Duplicate sends, charges, writes, or model calls would be costly.
42
43	Not usually worth it:
44
45	- One short, read-only model call with no side effects and no need for durable
46	progress.
47	- UI-only autocomplete where losing the request is acceptable.
48
49	## Architecture
50
51	Use this shape unless the repo already has a stronger established pattern:
52
53	1. The HTTP/server action layer validates auth, stores the user's intent if
54	needed, emits an event with a stable `id`, and returns quickly.
55	2. An Inngest function owns the agent run.
56	3. Load state and external context inside `step.run`.
57	4. Create AgentKit agents inside the function or import agent/network
58	factories.
59	5. Run model inference through AgentKit / `step.ai`; wrap non-model tool side
60	effects in `step.run`.
61	6. Use `step.waitForEvent` or `step.waitForSignal` for human approval and
62	external callbacks.
63	7. Publish durable progress with native realtime.
64	8. Add sessions and scores when the agent outcome needs to be evaluated later.
65	9. Apply flow control at the function level for provider and tenant limits.
66
67	## Basic AgentKit Function
68
69	Prefer a small, typed function first; add networks and extra tools after the
70	single-agent path is proven.
71
72	```typescript
73	import { createAgent, openai } from "@inngest/agent-kit";
74	import { inngest } from "@/inngest/client";
75
76	export const summarizeTicket = inngest.createFunction(
77	{
78	id: "summarize-ticket",
79	triggers: [{ event: "support/ticket.created" }],
80	concurrency: [{ key: "event.data.accountId", limit: 2 }]
81	},
82	async ({ event, step }) => {
83	const ticket = await step.run("load-ticket", () => {
84	return getTicket(event.data.ticketId);
85	});
86
87	const writer = createAgent({
88	name: "support-summary-writer",
89	system: "Write a concise support-ticket summary with next actions.",
90	model: openai({ model: "gpt-4o" })
91	});
92
93	const { output } = await writer.run(JSON.stringify(ticket));
94
95	await step.run("save-summary", () => {
96	return saveTicketSummary(event.data.ticketId, output);
97	});
98
99	return { ticketId: event.data.ticketId };
100	}
101	);
102	```
103
104	## Tool Calls
105
106	Tools can be defined with AgentKit, but agent-safe tools should still follow
107	durability rules:
108
109	- Read-only tool calls can run as part of the agent when replaying is harmless.
110	- External side effects should be isolated with stable IDs and `step.run`
111	boundaries, or implemented as tool handlers that use the provided `step`.
112	- Tool outputs should be small enough for step state limits.
113	- Validate tool parameters

People who liked this also viewed

6 picks

Type