Features
Every part,
designed together.
Caudra ships as one native binary, and every part on these pages comes with it. The first seven lead the way. The rest is everything else a modern coding agent should have.
What Caudra leads with
Autopilot
Answers prompts and questions while you are away, and logs every decision it makes.
Decisions and reranking
Screens web, MCP, and Python results, and ranks what the agent reads first.
Token savings
Filtered shell output, code outlines, fewer round-trips, and a prompt cache that holds.
Background subagents and shell calls
Subagents and long commands keep running while the agent moves on, and you can steer any of them.
Learner and memory
Corrections become instructions, and notes carry across sessions in a view that stays small.
Nudges
Small and flagship models get back on task when a turn stalls, loops, or stops early.
Code mode
MCP servers, past conversations, and decisions, reached through Python in one tool.
Everything else
Revert and history
Search every past session, revert files with a preview, and fork from any message.
Workbench
A file explorer, editor, project search, and source control beside the session.
Herdr and Orca
Every session reports its state to the host and opens forks and worktrees in tabs of their own.
Tools
File, shell, and web tools whose results say what they left out.
Models and setup
17 providers, local models, and the instruction files, skills, and MCP servers you already have.
Accounts and quota
Several subscriptions and keys per provider, and a conversation that moves before the quota runs out.
Privacy
Your history stays on your machine, and telemetry is off unless you run a collector.
Experimental lab
Durable workflows, VM sandboxes, and more, each off until you switch it on.
Designed together
The parts know
about each other.
The core team builds, tests, and releases every capability together, so one feature can hand work to the next with no extra setup. You install one binary, with no third-party plugins to vet or keep up to date. Lua extensions remain an experimental option.
- Workbench to Prompt
Ctrl+X Entersends the lines you selected to the composer as a mention such as@src/api.ts:L10-L20. - Shell to Transcript
The model reads the filtered output, and the footer switches you to the raw view.
- You to Running subagent
Type into a running task's chat, and the subagent reads it at its next turn boundary.
- Background work to
/goalThe goal check waits until background tasks and shell commands settle.
- Tool call to Revert
Each call that may change files is recorded, so a revert touches only the files those calls changed.
- Every model to
/usage/usageshows what the session spent and scores the cache hit rate of each model. - Spent account to
/modelWhen every account is spent, the card offers another model, and the waiting chat continues on it at once.
- Decision calls to Autopilot budget
Each decision call is priced in
/usageand counts toward the spend cap of autopilot and of a goal. - Background shell to Agent
Every 30 minutes a command runs, the agent waiting on it reads the latest output and decides whether to keep waiting.
Everything in the box
119 capabilities,
designed together.
Each one links to its documentation. Experimental items are marked and stay off until you switch them on.
Agents
28
- Steerable subagents
- Background tasks and shell jobs
- Shell commands that move to the background after 10 minutes
- A check on background shells that run long
- Guidance from the main agent to its tasks
- Switch the model of a running turn or task
- Next, Guide, and Replace queue
- Completion goals with a log of every check
- Goals that wait for CI and stop when stuck
- Autopilot toward a goal
- Replace a subagent turn from its chat
- Side questions with read-only tools, from
/btwandF2 - A summary of the session with
/catchup - Questions you can answer later
- Notes on your answers to the agent
- Task cards that list every subagent step
- A model for each job
- New models and prices from a signed list, without an update
- Several accounts per provider, switched by quota
- Another provider when every subscription runs out
- Tasks on the model you name
- Favorite models in
/model - Automatic steering, tuned per model
- Continual learning into
AGENTS.mdand memory - Requirements tracking with
/extract - Resume a stopped turn with
/continue - Work cut off by a crash resumes on reopen
- Durable workflowsExperimental
Tools
17
- File read and search, with one editor per model
- File outlines with
file_index - Source code graph for 38 languages and formats, TLA+ included
- Shell with output filtering
- Shell ETAs learned on your machine, shown in
/shells - Python for computation, history search, MCP, and modules loaded on demand
decisions.decide()anddecisions.rerank()inside snippets- JavaScript snippets translated to Python when the meaning holds
- Parallel calls with
batch - Retained output search
- Web search and fetch, with PDFs as text or as the file
- Web search through Exa, SearXNG, Brave, Kagi, or SerpApi
- Fetches from local and tailnet servers you list
- Image generation with a ChatGPT login
- Tool-call JSON repair
- Polling and waiting inside snippets
- Exact decimal arithmetic in Python
Context
15
- Instruction files with
AGENTS.md - Project memory with summaries
- Reminders to save the first notes in a project
- Skills from
.claudeand.agentsfolders - pstack engineering skills, off by default
- Matt Pocock's skills, off by default
- Emil Kowalski's design-engineering skills, off by default
@fileand#commitmentions- Inspect with
/contextand/projection - Compaction that keeps recent turns
- System prompt profiles
- Config edits apply while Caudra runs
- Config files that upgrade themselves, with backups
{env:NAME}secrets in global config- The full manual offline in
/docs
Sessions
19
- Revert conversation, files, or both
- File changes recorded per call
- Forks from any message, or the latest with
/fork - Forks in a new Herdr or Orca tab
- Tabs named after the session
- Git worktrees you open and close
- Retention policies and
/storage - Images and PDFs stored once across sessions
- Move sessions between directories
- Search this chat or past sessions with
/search - The model and thinking level behind each message
- Print runs saved as sessions you can resume
- Search past conversations from Python
- Spend by model and purpose in
/usage - Subscription spend beside billed spend
- Priced decision calls in
/usage, goals, and autopilot budgets - Opt-in cache warming for the Anthropic API
- Release notes after an update, with
/changelog - Update from the banner and restart on the same sessions
Interface
14
- Workbench with Git and search
- Passage review
- Markdown with maths, Mermaid flowcharts, and state diagrams
- Pictures in the transcript
- Tool calls drawn as they stream
- Token cost on tool cards
- Draft stash and command palette
- Commands that list their arguments and complete them
- A keybindings list you can filter
- A footer row for errors and notices
- Edit
caudra.tomlin/settings - Paired light and dark themes
- Hover tooltips that explain controls
- Notifications in tmux, Herdr, and Orca
Safety
11
- Parsed shell permissions
- Scoped approvals, down to a folder
- Patterns suggested from history
- Plan mode
- No tracking, opt-in telemetry
- Managed VM sandboxesExperimental
- Auto mode with decision screening
- Content screening that can withhold a result
- Every request through your proxy
- Secrets for shell commands on a private card
- JEV decision engine
Integrations
15
- MCP code mode: the model calls your servers from Python
- MCP OAuth with client ID metadata documents
- MCP results drawn by type, with token estimates
- MCP server questions and
@resources - ACP for Zed and desktop apps
- Session management and background subagents over ACP
- Headless runs for scripts and CI
- Claude Code compatible output
- Custom slash commands
- Herdr integration
- Orca integration
- Workcell tools, also a standalone MCP server
- Cross-session messagingExperimental
- Lua extensionsExperimental
- Remote Workcell workspacesExperimental
Why the name
Caudra comes
from caudate.
Caudra, pronounced KAW-druh, is named after the caudate nucleus. This part of the brain belongs to the circuits that connect evidence and goals to action. Studies link it to learning which actions lead to which outcomes.
The name describes how Caudra works. It turns the context of your task into an action, then reads the outcome before it chooses the next one.
The brain is only the inspiration for the name. Caudra is software, and it works toward the goal you set.
Sources: Grahn, Parkinson and Owen, 2008, Lau and Glimcher, 2007, Doi et al., 2020.
- 01Context and intent
- 02Evaluate evidence
- 03Select an action
- 04Execute with tools
- 05Observe the outcome