Tools

The tools behind DataMind Oda: subagents and delegation, commands, the browser, web access, background jobs, timeouts and more.

• Delegating work to subagents

  How Oda delegates work to subagents: the tool call, its parameters, foreground runs, and what the result looks like.

• Subagent settings

  Set how deep and how wide Oda may delegate: the recursion depth and parallelism limits, their defaults, and what overrides them.

• Background subagents and job collection

  Start one-shot background subagents, collect their output with the job tools, and stop them when they are no longer needed.

• Continuable subagents: follow-ups and control

  Follow up on a subagent: send messages, interrupt a turn, list children, deliver queue or steer prompts, and resume cold children.

• Subagent settlement and results

  Understand settlement notices, how a subagent reports back to its parent, and what a cold resume replays.

• Choosing the model for a subagent

  Turn on subagent model selection, discover allowed routes and efforts, and pass provider, model and reasoning effort.

• Choosing and enabling a delegation backend

  Choose among the delegation backends, enable one for a chat, and compare how they run, what they support and their limits.

• Delegation backend: ACP

  Configure and operate the ACP delegation backend: settings, permissions, stop reasons, diagnostics and run semantics.

• Delegation backend: Claude Code

  Configure and operate the Claude Code delegation backend: permission modes, unattended behaviour, limits and diagnostics.

• Delegation backend: Codex

  Configure and operate the Codex delegation backend: its three modes, wire identity, run semantics and limits.

• Subagent permissions and sandboxing

  How a delegated child inherits permissions: pinned approvals, no escalation, resume snapshots, and the unattended boundary.

• Structured subagent output

  Record a structured final answer from a subagent with the structured_output tool, its schema subset and error cases.

• Subagent failures and diagnostics

  Diagnose subagent runs: stop-reason mappings, per-backend stage and category vocabularies, refusals and failure formats.

• Command execution

  How commands run: spawn specs, environment scrubbing, output caps and spill, and executable resolution.

• Process containment and cleanup

  How Oda contains and cleans up processes: platform containment, terminal signalling, activity states and host-exit cleanup.

• Browser reference: tools, policy and limits

  Reference for the browser tools: parameters, bounds, error codes, element and tab references, policy and site grants.

• Plugins and bundles

  Manage plugins and bundles: toggle rows, install or remove bundles, approve build scripts, and read the errors and logs.

• Settings and credentials reference

  Reference for settings and credentials: write operations, conflict semantics, redaction, and the exact error codes.

• The profile settings file

  The per-computer settings file: its header, its backup, how patches are normalised, and how to disable it.

• Hot reload

  What applies live and what needs a restart: how Oda applies changes while it runs, how it watches for them, and what the reload messages mean.

• Command line and flags

  Start Oda with a profile, use flags beside the configuration, and read the config and schema dumps.

• Startup and diagnostics

  What happens when Oda starts, how to read its output when a start fails, where its environment values come from, and what needs a restart.

• Oda profiles and launchers

  How Oda runs in different modes: the desktop app, automation and SDK clients, and one-shot runs, plus adding bundles from the command line.

• Headless runs

  Run one task without opening the app: pass the task as text or stdin, stream the run as JSON events, or adopt an existing chat.

• The Oda web interface

  Open Oda's chat in a browser on this computer: launch the web mode, sign in from the printed link, and set host and port options.

• Chat compaction

  How Oda keeps long chats going: automatic compaction, the /compact command with its messages, and what a checkpoint keeps.

• Tool result pruning and image offload

  How Oda trims oversized tool results under pressure, and what happens to images sent to a model that runs out of room for them.

• Memory files, limits and history

  The memory files on this computer: format and versions, the limits that apply, and how memories reach the model.

• Workspace instructions

  Give Oda standing instructions for a folder with AGENTS.md files, and what Oda tells you when they change or stop applying.

• Time and terminal context

  The time and terminal context Oda reads while it works, and the settings that control when each reading is added.

• The agent task list

  The to-do list Oda keeps while it works: how it updates, the statuses it uses, and what the chat shows.

• Web search and page reading

  What Oda's web search and page reading can do: query and result limits, truncation, timeouts and the errors they can return.

• Direct web fetching

  The built-in page reader Oda can use without the platform: URL rules, network safety checks, redirects and limits.

• Timeouts in Oda

  The time budget behind every slow call: tool budgets, provider backstops, watchdog behaviour, and the messages each produces.

• Proxy support

  Use Oda behind a corporate proxy: the settings Oda reads, what is always bypassed, and what child programs receive.

• Where Oda keeps its data

  Oda's home folder: how it is chosen, how paths are shown, and how settings files are saved safely.

• File addresses

  How Oda names files in links and previews: chat-scoped and absolute addresses, and why some cannot be opened.

• Diagnostic text and logs

  What Oda rewrites in the technical text it shows or copies, what the rewrites never touch, and where the original wording stays.

• Runtime consistency checks

  The built-in checks that keep Oda's own state consistent: what they cover, the error they raise, and how they are configured.