Agent API

Pick any agent stack. One simple API.

Claude Code, Codex, Hermes, and every other supported harness behind a single stateful API. Choose the harness when you create a spec; keep the same session, message, and event interfaces in your integration.

Get API key Read the docs

Claude Code Codex Hermes OpenClaw pi DeepSeek Harness Kimi Code OpenCode

Agent API — https://api.agentsky.dev/v1

# 1 — Create a reusable agent spec
POST /v1/agents
{ "name": "pr-reviewer",
  "agentType": "codex",
  "llm": "gpt-5.6-sol" }

# Save response.agent.id (name is only a label).

# 2 — Start its cloud computer
POST /v1/sessions
{ "agent": "<response.agent.id>",
  "initial_events": [{
    "type": "user.message",
    "content": [{ "type": "text",
      "text": "Review PR #482" }]
  }] }

# Save response.session.id for the endpoints below.
# 3 — Continue: POST /v1/sessions/{id}/events
# 4 — Stream: GET /v1/sessions/{id}/stream

Agent sessions

Every session is a persistent cloud computer. State, workspace, and recovery are included on the same API call.

POST /v1/sessions

Models

Claude, GPT, DeepSeek, Kimi, GLM — choose the model that fits the job. Choose a compatible model when creating your agent spec.

POST /v1/agents · llm

Channels

Slack, Telegram, Discord, the web, and a direct API. Every session is addressable from the channel your users prefer.

POST /v1/channels/bindings

Connectors

Attach skills and tool APIs, then authorize the connected apps your agent needs.

POST /v1/agents · capabilities

Send a message

Send a message. Stream the reply.

Replace the example session ID with the ID returned at creation. Open the SSE stream, POST the message, then read events. These steps work for every supported harness.

POST /v1/sessions/{id}/events — send a turn: text, images and files

GET /v1/sessions/{id}/stream — SSE stream of events

GET /v1/sessions/{id}/events — paginated catch-up

Full message API reference

# Requires curl 7.76+
set -euo pipefail
: "${SKY_API_TOKEN:?Set SKY_API_TOKEN first}"

BASE_URL='https://api.agentsky.dev/v1'
SESSION_ID='sess_example123'
PAYLOAD='{"events":[{"type":"user.message","content":[{"type":"text","text":"Review the diff in the current workspace."}]}]}'

# Open the SSE stream first so no reply event is missed.
(
  sleep 0.5
  curl --fail-with-body --silent --show-error \
    --request POST "$BASE_URL/sessions/$SESSION_ID/events" \
    --header "Authorization: Bearer $SKY_API_TOKEN" \
    --header "Content-Type: application/json" \
    --data "$PAYLOAD" >&2
) &

curl --fail-with-body --silent --show-error --no-buffer \
  "$BASE_URL/sessions/$SESSION_ID/stream" \
  --header "Authorization: Bearer $SKY_API_TOKEN"

Supported agents

Every harness. One contract.

Choose a harness and compatible model when creating an agent spec. To try another harness, create a different spec and start a session from its returned ID.

Model pricing →

Agent / Harness	Compatible models	Context	Model billing
Claude Code	Claude Opus 5 · Claude Fable 5.1 · Claude Sonnet 5 · Claude Haiku 4.5	Varies by model	per token or eligible BYO subscription
Codex	GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · GPT-6 Astra · DeepSeek V4 Pro · DeepSeek V4.1 Flash	Varies by model	per token or eligible BYO subscription
Hermes	DeepSeek V4 Pro · DeepSeek V4.1 Flash · GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · GPT-6 Astra · Gemini 3.8 Flash · GLM-5.3 · GLM-5.3-Flash · Kimi K3 · Kimi K2.7 Code · Grok 4.6 · Claude Opus 5 · Claude Fable 5.1 · Claude Sonnet 5 · Claude Haiku 4.5	Varies by model	per token
OpenClaw	GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · GPT-6 Astra · DeepSeek V4 Pro · DeepSeek V4.1 Flash · Gemini 3.8 Flash · GLM-5.3 · GLM-5.3-Flash · Kimi K3 · Kimi K2.7 Code · Grok 4.6 · Claude Opus 5 · Claude Fable 5.1 · Claude Sonnet 5 · Claude Haiku 4.5	Varies by model	per token
pi	GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · GPT-6 Astra · DeepSeek V4 Pro · DeepSeek V4.1 Flash · GLM-5.3 · GLM-5.3-Flash · Kimi K3 · Kimi K2.7 Code · Grok 4.6 · Claude Opus 5 · Claude Fable 5.1 · Claude Sonnet 5 · Claude Haiku 4.5	Varies by model	per token
DeepSeek Harness	DeepSeek V4 Pro · DeepSeek V4.1 Flash · Kimi K3 · Kimi K2.7 Code · GLM-5.3 · GLM-5.3-Flash · Grok 4.6	Varies by model	per token
Kimi Code	Kimi K3 · Kimi K2.7 Code · DeepSeek V4 Pro · DeepSeek V4.1 Flash · GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · GPT-6 Astra · GLM-5.3 · GLM-5.3-Flash · Grok 4.6 · Claude Opus 5 · Claude Fable 5.1 · Claude Sonnet 5 · Claude Haiku 4.5	Varies by model	per token
OpenCode	GPT-5.6 Sol · GPT-5.6 Terra · GPT-5.6 Luna · GPT-6 Astra · DeepSeek V4 Pro · DeepSeek V4.1 Flash · GLM-5.3 · GLM-5.3-Flash · Kimi K3 · Kimi K2.7 Code · Grok 4.6 · Claude Opus 5 · Claude Fable 5.1 · Claude Sonnet 5 · Claude Haiku 4.5	Varies by model	per token

Active compute is billed separately from model tokens or your eligible subscription. See pricing

Multi-agent flows

One task, many agents working in sequence.

Create a spec for each role, then start its session. Your orchestrator reads the planning session's events and sends the selected plan to the coding session; it passes the resulting code or repository reference to the testing session. The same message and event contract connects every step.

Try it now →

# Create each role with POST /v1/agents
# Save each response's agent.id (names are labels).
{"name":"planner","agentType":"claude_code"}
{"name":"coder","agentType":"codex"}
{"name":"tester","agentType":"hermes"}

# After reading the planner's /events, send its
# output to the coder. Replace both placeholders.
POST https://api.agentsky.dev/v1/sessions
{
  "agent": "<coder agent.id>",
  "initial_events": [{
    "type": "user.message",
    "parts": [{"type":"text","index":0,
      "text":"<plan read from the planner session>"}]
  }]
}
# Read coder events; pass its result to the tester.

Your own product

Give your users an agent, not a chat window.

Create a reusable spec with your product's instructions and tools. Save the returned agent ID, then create a separate session for each user. Your backend keeps each session ID and sends subsequent messages to that same session.

Try it now →

# Create your product's agent spec
POST https://api.agentsky.dev/v1/agents
{
  "name": "support-agent",
  "agentType": "claude_code",
  "prompt": "You are the support agent for Acme..."
}
# Save response.agent.id; name is not an ID.

# Each user gets their own session
POST https://api.agentsky.dev/v1/sessions
{
  "agent": "<response.agent.id>",
  "metadata": {"userId":"<your user ID>"}
}
# Save response.session.id for messages and events.

Built with the API

Real software factories running in production.

Tycoon

AI company operating system — 150,000 agents launched through one API.

Read the story →

HeyBoss AI

Software-creation product — 250,000 agents handling full build-to-deploy pipelines.

Read the story →

WebJourney

Research tool — 50,000+ web-research sessions run autonomously.

Read the story →

Pricing

List price, per token. $0 while parked.

Compute is metered by the second — only while a session is active. Model tokens are billed at list price. Connect your own Claude or OpenAI subscription to pay $0 for eligible model usage.

See full pricing

Get your API key. Build today.

One key for the supported harnesses. Create a spec, start a session, and send your first task.

Get API key Explore the docs