You're looking at a specific version of this model. Jump to the model overview.

fmxexpress /pasclaw-build-deploy:3b759172

Input schema

The fields you can use to run this model with an API. If you don’t give a value for a field its default value will be used.

Field Type Default value Description
message
string
Build task. The agent runs up to max_iters " "tool-using iterations to satisfy it.
max_iters
integer
50

Min: 1

Max: 500

Tool-loop iteration budget. One iteration ~= one " "model response (which may include multiple tool calls).
timeout_seconds
integer
3600

Min: 30

Subprocess timeout in seconds. Default 3600 = 1 h. " "Set high for long horizon builds; Replicate enforces its own " "container ceiling on top of this.
workspace_in
string
Workspace archive from a previous build. " "Leave empty for a fresh run. Replicate handles both " "file-uploads and URL inputs through this. For very " "large URL-backed workspaces, prefer workspace_in_url " "to get parallel pget download.
workspace_in_url
string
Explicit URL to a previous-build workspace " "archive. Leave empty for a fresh run. When set, " "downloaded via pget for parallelism; takes precedence " "over workspace_in.
openai_api_key
string
OpenAI API key. Stored encrypted; never surfaces in logs.
anthropic_api_key
string
Anthropic API key. Stored encrypted; never surfaces in logs.
gemini_api_key
string
Google Gemini API key. Stored encrypted; never surfaces in logs.
groq_api_key
string
Groq API key. Stored encrypted; never surfaces in logs.
openrouter_api_key
string
OpenRouter API key. Stored encrypted; never surfaces in logs.
deepseek_api_key
string
DeepSeek API key. Stored encrypted; never surfaces in logs.
ollama_url
string
Base URL of an Ollama server reachable from the cog " "container (e.g. an ngrok tunnel to a laptop's local " "instance). Leave empty to skip.
lmstudio_url
string
Base URL of an LM Studio server reachable from the " "cog container (default LM Studio port is 1234). Leave " "empty to skip.
vllm_url
string
Base URL of a vLLM server reachable from the cog " "container (default vLLM port is 8000). Leave empty to skip.
custom_provider_kind
string
Catalog kind for a custom provider (e.g. mistral, " "xai, cerebras, moonshot, qwen, zhipu, perplexity, " "nvidia, volcengine, minimax, novita, litellm, mimo). " "Use 'openai-compat' for any other OpenAI-compatible " "endpoint not in PasClaw's catalog. Leave empty to skip.
custom_provider_url
string
api_base URL for the custom provider. Required when " "custom_provider_kind is set.
custom_provider_key
string
API key for the custom provider. Stored encrypted; " "never surfaces in logs. Leave empty if the endpoint " "needs no auth.
custom_provider_model
string
Default model id for the custom provider. Leave " "empty to let the provider pick (some servers route " "missing model ids to whatever is currently loaded).
provider
string
LLM provider to route through. One of openai, " "anthropic, gemini, groq, openrouter, deepseek, ollama, " "lmstudio, vllm, or whatever you put in " "custom_provider_kind. Empty = first configured " "provider wins.
model
string
Override model id. Empty = provider's catalog default.
mode
None
build
None
profile
string
max-build
PasClaw config profile applied on top of stock " "defaults. `max-build` (default) turns on web_fetch, " "vector_search, auto_router, prompt_cache, task-aware " "memory orientation, promptware guard, the self-improving-" "skills suite, and bumps tool_output_cap to 16 KB -- the " "richest unattended-build toolset. (Note: web_search is a " "max-build *flag*, but the tool only registers when " "PasClaw is also configured with a search provider. This " "cog doesn't supply one, so web_search isn't available " "out of the box -- use web_fetch.) Other choices: " "`baseline` (everything off, useful for A/B), `low-token` " "(cheaper, smaller context window), `security` (workspace " "restriction + shell deny + private-network block), " "`all-on` (max-build plus every remaining flag flipped " "on). Empty to skip profile application.
deploy_to_cloudflare
boolean
True
After pasclaw build completes, run `wrangler " "deploy --temporary` against any wrangler.toml the agent " "produced in the workspace. Cloudflare's temporary-accounts " "feature (https://blog.cloudflare.com/temporary-accounts/) " "gives you a 60-minute throwaway account + a claim URL to " "convert it to permanent, with no API key required on the " "cog side. Skipped silently if no wrangler.toml is in the " "workspace. Set False to skip the deploy step entirely.
wrangler_project_path
string
Path INSIDE the workspace to the directory " "containing wrangler.toml. Default empty = auto-detect " "the first wrangler.toml found under " "$PASCLAW_HOME/workspace/. Use this when the agent " "produces multiple wrangler projects and you want a " "specific one deployed.

Output schema

The shape of the response you’ll get when you run this model with an API.

Schema
{'items': {'format': 'uri', 'type': 'string'},
 'title': 'Output',
 'type': 'array'}