1. Create an API key
Create or sign in to your IteraCompute account, then create a key in the Console.
Connect a terminal agent to IteraCompute.
Create or sign in to your IteraCompute account, then create a key in the Console.
Configure a named custom provider with chat_completions. The Python AIAgent and CLI configurations are separate profiles.
Use the exact API Model ID for your chosen model. Check its context and capabilities before enabling agent or vision features.
DeepSeek V4.1 Flash 1017 is the paid reference model. You can replace it with a published conversation model allowed by your key. Recheck model permissions, prices, context and tool support after switching; the examples follow your selection, while the retained test results remain tied to Qwen.
Use an isolated Hermes Agent v0.23.0 checkout and HERMES_HOME. Export ITERACOMPUTE_API_KEY in the launching terminal.
git clone --branch v0.23.0 --depth 1 \
https://github.com/NousResearch/hermes-agent.git hermes-agent-v0.23.0
cd hermes-agent-v0.23.0
test "$(git rev-parse HEAD)" = \
"2bd378ed1b8d6bf2192cb26e0a55ff3091e6bd19"
export HERMES_HOME="$PWD/.hermes-iteracompute-cli"
export HERMES_RUNTIME_DIR="$HERMES_HOME/tools"
umask 077
mkdir -p "$HERMES_HOME"
source ./activate
hermes --version
hermes pm doctorumask 077
read -rsp 'IteraCompute API key: ' ITERACOMPUTE_API_KEY
printf '\n'
export ITERACOMPUTE_API_KEYHermes CLI may add reasoning_effort=medium for an unknown custom model. Confirm the actual request against the selected model parameters. reasoning_config={} below is a Python constructor option, not a YAML setting or proof that CLI reasoning is disabled.
Keep API keys in your server environment or tool credential store. Replace the example key field with your own key where applicable.
providers:
iteracompute:
api: https://api.iteracompute.com/v1
key_env: ITERACOMPUTE_API_KEY
transport: chat_completions
default_model: deepseek/deepseek-v4.1-flash-1017
context_length: 1048576
discover_models: false
extra_body:
max_tokens: 512
model:
provider: custom:iteracompute
default: deepseek/deepseek-v4.1-flash-1017
context_length: 1048576
streaming: false
agent:
max_turns: 3
api_max_retries: 0
auto_recovery_cycles: 0
fallback_providers: []
toolsets: [todo]
tools:
tool_search:
enabled: off
approvals:
mode: manual
compression:
enabled: false
auxiliary:
title_generation:
enabled: false
model_upgrade_enabled: falseSave this configuration only in the isolated profile created above. These Chat Completions examples start with 512 output tokens and no fallback. Context metadata is not an account entitlement; OpenClaw uses a smaller input budget. The Qwen selection uses 128K account context. Reasoning controls vary by model: the Python and OpenClaw examples add reasoning: false only for Qwen; do not transfer that option to other models.
python3 - <<'PY'
import json
import os
import urllib.error
import urllib.request
payload = {
'model': 'deepseek/deepseek-v4.1-flash-1017',
'messages': [{'role': 'user', 'content': 'Reply with exactly: ITERACOMPUTE_API_OK'}],
'max_tokens': 512,
**{},
}
request = urllib.request.Request(
'https://api.iteracompute.com/v1/chat/completions',
data=json.dumps(payload).encode('utf-8'),
headers={'Authorization': 'Bearer ' + os.environ['ITERACOMPUTE_API_KEY'],
'Content-Type': 'application/json'}, method='POST',
)
try:
with urllib.request.urlopen(request, timeout=180) as response:
result = json.load(response)
except urllib.error.HTTPError as exc:
raise SystemExit(f'HTTP {exc.code}; inspect redacted details privately') from None
except (urllib.error.URLError, TimeoutError):
raise SystemExit('Network or timeout failure; no automatic retry') from None
choice = result['choices'][0]
if ((choice.get('message', {}).get('content') or '').strip() != 'ITERACOMPUTE_API_OK'
or choice.get('finish_reason') != 'stop'):
raise SystemExit('FAIL: expected final marker and stop reason not returned')
print('Direct API text passed; test the agent separately')
PYThis sends one real request when you run it. Confirm the exact visible marker and finish_reason=stop; an empty HTTP 200 response is incomplete. Paid model calls consume API balance. Do not automatically retry an unknown outcome.
hermes chat --oneshot --ignore-rules --toolsets todo --max-turns 2 \
--format stream-json \
-q 'Do not call tools. Reply with exactly: HERMES_ITERACOMPUTE_OK' \
> hermes-text.private.jsonl 2> hermes-text.private.stderrRun the direct API check below first, then require a non-empty final marker, the intended provider/model and a clean exit from the agent. Configuration validation and a provider probe do not prove an agent or tool workflow.
The CLI profile above is a documentation reference. For the Python route, create a different HERMES_HOME inside the activated checkout, save the following config.yaml there, and save the Python script as iteracompute_agent_smoke.py. Do not combine the two profiles.
export HERMES_HOME="$PWD/.hermes-iteracompute-python"
mkdir -p "$HERMES_HOME"model:
provider: custom
default: deepseek/deepseek-v4.1-flash-1017
base_url: https://api.iteracompute.com/v1
streaming: false
providers:
custom:
request_timeout_seconds: 240
stale_timeout_seconds: 240
tools:
tool_search:
enabled: off
agent:
api_max_retries: 0
auto_recovery_cycles: 0
fallback_providers: []
compression:
enabled: false
auxiliary:
title_generation:
enabled: false
model_upgrade_enabled: falseimport os
from run_agent import AIAgent
agent = AIAgent(
provider="custom",
api_mode="chat_completions",
base_url="https://api.iteracompute.com/v1",
api_key=os.environ["ITERACOMPUTE_API_KEY"],
model="deepseek/deepseek-v4.1-flash-1017",
enabled_toolsets=["todo"],
max_iterations=3,
max_tokens=512,
run_budget_seconds=600,
reasoning_config={},
request_overrides={"extra_body": {}},
skip_context_files=True,
skip_memory=True,
skip_background_review=True,
load_soul_identity=False,
save_trajectories=False,
quiet_mode=True,
fallback_model={},
)
try:
result = agent.run_conversation(
"Call todo_list exactly once to write one item with id smoke-1, "
"content API smoke test, and status completed. Then read the tool result "
"and reply with exactly HERMES_ITERACOMPUTE_TOOL_OK. Do not use other tools."
)
if not (result.get("completed") is True and
(result.get("final_response") or "").strip() == "HERMES_ITERACOMPUTE_TOOL_OK"):
raise SystemExit("FAIL: inspect redacted details privately")
print("Final marker passed; verify actual todo_list events separately")
finally:
agent.close()python iteracompute_agent_smoke.pySuccess criterion: inspect a real todo_list call, its completed smoke-1 item, the returned tool result and the final HERMES_ITERACOMPUTE_TOOL_OK reply. The script checks the final marker only; tool events need separate inspection. Three iterations and zero configured retries are not a strict HTTP-request or spending cap.
The model API runs remotely; no local GPU or public Gateway port is needed. The agent still needs local CPU, memory and permissions. Keep keys in the launching process environment, keep raw logs private, and enable only the tools needed for your test. External tools and services have their own setup and costs.
# Bash / zsh / WSL; set ITERACOMPUTE_API_KEY in your environment first.
curl --fail-with-body https://api.iteracompute.com/v1/chat/completions \
-H "Authorization: Bearer $ITERACOMPUTE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"deepseek/deepseek-v4.1-flash-1017","messages":[{"role":"user","content":"Reply with exactly: Hello!"}],"max_tokens":512}'Check that choices[0].message.content contains visible text; HTTP 200 alone is insufficient. Reasoning may consume the output budget. For free Auto, use the first free call example to disable thinking, then send a prompt inside your tool.
Capabilities describe the IteraCompute API. Feature availability in your workflow also depends on the tool and its settings; reasoning controls vary by model.
Check the launching process key, model permission and paid balance. Keep /v1 once in the base URL. Send the raw Model ID to the API; the iteracompute/ prefix belongs only to OpenClaw routing.
Check model-specific reasoning parameters, output budget and account limits. Honor Retry-After. Do not stack reasoning dialects, infer success from HTTP 200 or switch to another paid model automatically.
Hermes: disable tool_search for the todo test and keep Python and CLI profiles separate. OpenClaw: replace deny-all with read-only for the tool test; do not use --isolated or --auth-env-only with this custom config.
Connect your key, choose a model and send your first request.