Agents articles
Updated
Running coding agents and parallel agent fleets on an OpenAI-compatible API: sessions, concurrency, rate limits and long context. 16 articles, each opening with a direct answer. Every outside fact links to its source with the date we checked it.
Claude Code with open models: proxies, native endpoints and limits
Claude Code sends Anthropic Messages API requests to ANTHROPIC_BASE_URL. To use an OpenAI-compatible provider, run a translating proxy such as LiteLLM and point Claude Code at it, or pick a provider with its own Anthropic-format endpoint, such as DeepSeek or Z.ai. Anthropic does not support non-Claude models in Claude Code, so test tools, caching and context limits yourself.
OpenClaw and Hermes Agent on a cheap model: setup and token math
Both agents accept a custom OpenAI-compatible endpoint: OpenClaw under models.providers with api "openai-completions", Hermes Agent with provider: custom and a base_url. Cost is driven by context size times how often the agent runs times the model price. Shrink the context, run scheduled checks less often and on a cheaper model, and test tool calling before you commit.
Prompt injection in coding agents: defences that actually hold
Assume a coding agent will sometimes follow instructions hidden in a README, issue, web page or tool result; OWASP says fool-proof prevention may not exist. So limit what a hijacked agent can do: sandbox it with no network by default, keep secrets out, allow-list commands, require approval to push or install, and review every change.
What one AI coding agent run costs: steps, context growth, caching
An agent resends its whole context on every step, so input grows with the square of the step count: 60 steps cost about 3.5 times the input of 30. Caching decides most of the bill; with append-only history, over 90% of input can bill at the cached rate. Multiply cost per run by monthly runs and compare with a flat session.
Agent token budget: where an agent loop's tokens go
An agent loop resends its whole context on every call: system prompt, tool definitions, conversation history, file contents and tool output, plus the new step. Input usually outweighs output many times over. Log prompt_tokens, cached_tokens and completion_tokens per call, find the biggest repeating block, then trim tools, cap tool output and restart long sessions.
Best LLM for Aider: how to choose a model and connect it
The best model for Aider follows its edit format reliably, holds your repository map in context and is cheap enough to run all day. Aider connects to any OpenAI-compatible API: set OPENAI_API_BASE to https://inference.broadnet.ai/v1, set OPENAI_API_KEY, and start aider --model openai/broadnet/deepseek-v4.1-flash. On a Full Moon plan that model has no token meter.
Best open model for coding agents: pick by tools, context and cost
For a coding agent that depends on native tool calls (opencode, Continue Agent mode, Zoo Code), GLM 5.3 is the safest pick on BroadNet: tools listed, 1M context. For Aider or an all-day agent, DeepSeek V4.1 Flash on a flat plan costs least. Qwen3.8 27B fits screenshot work. Gemma 4 VL 26B's 32K context is too small.
Context management for long-running agents with a 1M window
Treat the context as an append-only log: a fixed system prompt and tool list first, then history that only grows, so each turn reuses the cached prefix. With a 1,048,576-token window, as on DeepSeek V4.1 Flash, compact rarely and in large steps, trim bulky tool output first, and keep durable facts in a short notes file the agent rereads.
How to run coding agents unattended overnight, safely
Run the agent in a container or VM with only the repository mounted, on its own git branch, committing after every completed step. Give it a written task file, a step cap and a wall-clock deadline, then review the branch in the morning. A flat plan session, such as BroadNet's $60 Full Moon, fixes the worst-case bill.
How to run parallel coding agents on a flat-rate LLM API
Give each agent that works at the same time its own plan session. A session runs one request at a time with no token meter, so an orchestrator with three sub-agents needs four sessions. Cap requests in flight in your own code to match. One key can hold several sessions, and each session on Full Moon costs $60 a month.
Long-running agents: timeouts, retries and checkpoints that hold
Treat each model call as disposable and the agent's state as durable. Set a timeout per call, stream to detect stalls, retry 429, 5xx and dropped streams with bounded backoff, and give every side-effecting tool an idempotency key. Save the message list after each step so a crash resumes from the last checkpoint, not from the start.
n8n AI Agent on a cheap LLM: setup, model choice and cost per run
Point n8n's OpenAI credential at a low-cost OpenAI-compatible API, enter the model by ID and turn off Use Responses API. Give the AI Agent a model with tool calling, such as Qwen3.8 27B or GLM 5.3. Keep Max Iterations low, memory short and tools few, and route simple classification to a cheaper model first.
Use MCP servers with OpenAI-compatible models (any provider)
MCP is model-independent. The MCP client in your app lists each server's tools, passes them to the model as ordinary function-calling tools, and runs tools/call when the model asks. Any OpenAI-compatible model with reliable tool calling can use MCP servers. Configure them in Cline, Continue, opencode or n8n, or bridge them in about 40 lines of Python.
Use open models in coding agents: Aider, Cline, opencode, Codex
Every major coding agent takes three values for a custom model: a base URL, an API key and a model ID. Aider uses OPENAI_API_BASE, Cline and Roo Code an OpenAI Compatible provider, opencode a custom provider, Continue apiBase. Codex CLI now speaks only the Responses API, so a chat-completions endpoint needs a translating proxy such as LiteLLM.
What "unlimited" really means in an LLM API plan
"Unlimited" in an AI plan can still hide a token meter, a rolling usage window, a weekly cap, a rate limit or a concurrency limit. Ask which of these apply. BroadNet's Full Moon plan has no token meter, no 5-hour window and no weekly cap for DeepSeek V4.1 Flash; its one limit is one request at a time per session.
What multi-agent setups really cost: per token vs flat sessions
A multi-agent setup costs roughly the tokens of every agent added together, plus handoffs and repeated context. Anthropic measured about 15 times the tokens of a chat for its multi-agent research system, against 4 times for one agent. Per token, the bill grows with every agent; on a flat plan it grows with the agents running at once.






