Skip to main content

Agent Runtime 🤖

WednesdayAI runs a single embedded agent runtime derived from pi-mono.

Workspace (required)

WednesdayAI uses a single agent workspace directory (agents.defaults.workspace) as the agent’s only working directory (cwd) for tools and context. Recommended: use wednesdayai setup to create ~/.openclaw/openclaw.json if missing and initialize the workspace files. Full workspace layout + backup guide: Agent workspace If agents.defaults.sandbox is enabled, non-main sessions can override this with per-session workspaces under agents.defaults.sandbox.workspaceRoot (see Gateway configuration).

Bootstrap files (injected)

Inside agents.defaults.workspace, WednesdayAI expects these user-editable files:
  • AGENTS.md — operating instructions + “memory”
  • SOUL.md — persona, boundaries, tone
  • TOOLS.md — user-maintained tool notes (e.g. imsg, sag, conventions)
  • BOOTSTRAP.md — one-time first-run ritual (deleted after completion)
  • IDENTITY.md — agent name/vibe/emoji
  • USER.md — user profile + preferred address
On the first turn of a new session, WednesdayAI injects the contents of these files directly into the agent context. Blank files are skipped. Large files are trimmed and truncated with a marker so prompts stay lean (read the file for full content). If a file is missing, WednesdayAI injects a single “missing file” marker line (and wednesdayai setup will create a safe default template). An absent BOOTSTRAP.md is the exception: it is omitted from the injected context with no marker, because it is supposed to be gone once onboarding completes. BOOTSTRAP.md is only created for a brand new workspace (no other bootstrap files present). The agent finishes the ritual by deleting it with the built-in delete tool, and after that it is not recreated and no longer injected on later restarts. To disable bootstrap file creation entirely (for pre-seeded workspaces), set:

Built-in tools

Core tools (read/exec/edit/write/delete and related system tools) are always available, subject to tool policy. delete removes one file per call and is always confined to the agent workspace (or the lane workspace when lanes are on), even when tools.fs.workspaceOnly is false. apply_patch is optional and gated by tools.exec.applyPatch. TOOLS.md does not control which tools exist; it’s guidance for how you want them used.

Tool profiles

Tool profiles set a baseline tool set before allow/deny lists apply. Set globally with tools.profile or per-agent with agents.list[].tools.profile:
  • minimal — read, write, edit, exec only
  • coding — coding-oriented tools (read, write, edit, delete, exec, apply_patch, browser)
  • messaging — messaging-oriented tools (message, sessions tools). Note: this profile has no filesystem tools, so onboarding adds tools.alsoAllow: ["group:fs"] on new deployments to give the agent workspace-contained read/write/edit/delete
  • full — all tools (default when no profile is set)

Tool allow/deny

Fine-grained control via tools.allow, tools.deny, and per-agent agents.list[].tools.allow/deny. Use tools.alsoAllow for additive entries that merge into the existing allowlist without replacing it. Deny always wins. The gateway tool also matches action tokens (gateway.config_apply, gateway.config_patch, and so on). For non-sandbox agents, whole-document config.apply is denied by default. Re-enable with tools.allow or tools.alsoAllow (including agents.list[].tools.alsoAllow). Bare * does not count. See Config writes. Per-provider overrides: tools.byProvider.<provider>.allow/deny scopes tool policy to a specific model provider or provider/model pair.

Session tool visibility

tools.sessions.visibility controls which sessions session tools (sessions_list, sessions_history, sessions_send, agents_nudge) can target:
  • self — only the current session
  • tree — current session + spawned subagent sessions (default)
  • agent — any session belonging to the current agent id
  • all — any session (cross-agent still requires tools.agentToAgent)

Agent-to-agent messaging

tools.agentToAgent controls cross-agent session tools. Disabled by default:

Sub-agent tool policy

tools.subagents.tools sets the default tool policy for spawned sub-agents. See Sub-agents for full details.

Filesystem restrictions

tools.fs.workspaceOnly: true restricts filesystem tools (read/write/edit/apply_patch) to the agent workspace directory. Unset (the default), read, write and edit can reach any path the gateway user can access; the sandbox is also off by default. apply_patch stays workspace-only unless tools.exec.applyPatch.workspaceOnly is also set to false. The delete tool is always workspace-contained regardless of this setting. Set workspaceOnly: true as a single guardrail against accidental writes outside the workspace. Known limitation: with workspaceOnly: true, read refuses bundled and managed SKILL.md paths outside the workspace, so skill instructions cannot be loaded. Workaround: set tools.fs.workspaceOnly: false.

Exec configuration

tools.exec controls exec tool behavior: host (sandbox/gateway/node), security (deny/allowlist/full), ask (off/on-miss/always), safeBins, and more. Per-agent overrides via agents.list[].tools.exec.

Loop detection

tools.loopDetection guards against repetitive tool-call patterns. See Tool-loop detection.

Skills

WednesdayAI loads skills from three locations (workspace wins on name conflict):
  • Bundled (shipped with the install)
  • Managed/local: ~/.openclaw/skills
  • Workspace: <workspace>/skills
Skills can be gated by config/env (see skills in Gateway configuration).

pi-mono integration

WednesdayAI reuses pieces of the pi-mono codebase (models/tools), but session management, discovery, and tool wiring are WednesdayAI-owned.
  • No pi-coding agent runtime.
  • No ~/.pi/agent or <workspace>/.pi settings are consulted.

Sessions

Session transcripts are stored in the SQL session database:
  • ~/.openclaw/sessions.db (SQLite default; per-agent index identity at ~/.openclaw/agents/<agentId>/sessions/sessions.json)
The session ID is stable and chosen by WednesdayAI. Legacy Pi/Tau session folders are not read.

Steering while streaming

When queue mode is steer, inbound messages are injected into the current run. The queue is checked after each tool call; if a queued message is present, remaining tool calls from the current assistant message are skipped (error tool results with “Skipped due to queued user message.”), then the queued user message is injected before the next assistant response. When queue mode is followup or collect, inbound messages are held until the current turn ends, then a new agent turn starts with the queued payloads. See Queue for mode + debounce/cap behavior. Block streaming sends completed assistant blocks as soon as they finish; it is off by default (agents.defaults.blockStreamingDefault: "off"). Tune the boundary via agents.defaults.blockStreamingBreak (text_end vs message_end; defaults to text_end). Control soft block chunking with agents.defaults.blockStreamingChunk (defaults to 800–1200 chars; prefers paragraph breaks, then newlines; sentences last). Coalesce streamed chunks with agents.defaults.blockStreamingCoalesce to reduce single-line spam (idle-based merging before send). Non-Telegram channels require explicit *.blockStreaming: true to enable block replies. Verbose tool summaries are emitted at tool start (no debounce); Control UI streams tool output via agent events when available. More details: Streaming + chunking.

Model refs

Model refs in config (for example agents.defaults.model and agents.defaults.models) are parsed by splitting on the first /.
  • Use provider/model when configuring models.
  • If the model ID itself contains / (OpenRouter-style), include the provider prefix (example: openrouter/moonshotai/kimi-k2).
  • If you omit the provider, WednesdayAI treats the input as an alias or a model for the default provider (only works when there is no / in the model ID).

Configuration (minimal)

At minimum, set:
  • agents.defaults.workspace
  • channels.whatsapp.allowFrom (strongly recommended)

Next: Group Chats 🦞