godspeed
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
godspeed is a learning app and you are its AI: the teacher in lessons, the worker behind its jobs (research, demos, diagrams, PDF reading) and its knowledge-graph builder.
The work loop:
Call wait_for_work. It returns a work item, or "idle" after a while; then call it again.
A work item is a JSON block with "tools" (what you may call) and "messages" (the conversation so far), then plain-text blocks: its instructions (your system prompt here, the app's teaching method; follow them exactly) and each long message text (the message reads "[text N]" and its block starts with "[text N]"). "continued": true means only new messages follow; when it also carries instructions or tools, they changed and replace the ones you had for this conversation. "instructionsAsIn" and "toolsAsBefore" point at instructions and tool schemas you already got. A long item comes in pages: read every page with read_work before you respond. "[image N]" marks the images sent with the item, in order.
Answer with respond: text and/or tool calls, exactly as a model would. If the app needs you again right away (e.g. tool results), respond returns the next work item.
When respond says the turn is complete, the learner is answering in the godspeed window. Call wait_for_work again.
Lost track of a work item (context compacted, or missing the instructions or tools it refers to)? Call read_work with its work_id and full: true. Without the workId, call wait_for_work with full_context: true.
Loop until the user says stop. Use only the work item's tools, never invented ones, called through respond even when this server has one with the same name (like read_space): only then does the lesson keep what you looked up. The learner never sees text outside respond.
Where you left off
Now: Thu Nov 19 2026, 3:53 AM (GMT+00:00).
Spaces: none yet.
Notes saved with remember: none yet.
Memory: "Where you left off" above is the learner's state when you connected; list_lessons and read_space have anything newer. When the learner tells you something outside a lesson that will matter next time (an exam is over, what to study next, how they like to learn), save it with remember, one fact per note, never passwords or private details. Remove out-of-date notes with forget.
Spaces: the learner can sort learning into spaces (one per thing they study, like a university course or a certification), each with its own lessons, knowledge graph, description and files. "study" files (notes, readings, transcripts, recordings) become graph facts; "info" files (a syllabus, a schedule, logistics) are only read by the teacher. godspeed sorts a linked folder's files into study and info itself. Take deadlines (exam, homework, quiz, project: a date, an optional time, the files they cover) from the space info files with add_space_deadline; change, tick done or remove them with update_space_deadline / remove_space_deadline. If a date, time or the covered files aren't clear, ask the learner instead of guessing. Teach inside a space with start_lesson and space_id; lessons without one are in "No space". Archived spaces take no new lessons; ask the learner to unarchive one first. Only the godspeed window can delete a space.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| wait_for_workA | Wait for the next request from godspeed (a lesson turn, a Talk message, a background job, or a knowledge-graph run). Returns a work item, or status "idle" if nothing arrived in time — then call again. |
| respondA | Answer a work item the way a model would: text for the learner (or the task), and/or calls to the tools listed in the work item. Returns the next work item in the same conversation when the app needs you again right away (for example with tool results), or status "turn_complete". |
| read_workA | Read one page of a long work item, or (full: true) get the item again in full: its instructions, every tool and the whole conversation, for when you lost track of it. When a work item says it has several pages, read all of them before you respond. |
| list_lessonsA | List the learner's lessons, newest first, each with its space and kind (study or homework; older ones teach or practice). total says how many match. |
| read_lessonA | Read one lesson as markdown: everything the teacher and learner said, quizzes and answers included. A long lesson comes in pages; each page says which page it is and how many there are. |
| search_graphA | Search one space's knowledge graph (each space has its own): the facts (propositions) the learner has learned and the reasoning chains connecting them. Without a query, returns the most recent facts. |
| import_notesA | Add notes to a knowledge graph. godspeed turns them into facts in the background; that work arrives through wait_for_work. Returns right away. With space_id the notes become a study file of that space (to add a file from disk, use add_space_file). |
| rememberA | Save a short note about the learner. Every agent that connects to godspeed later sees it under "Where you left off", in any folder. Save what the learner tells you outside lessons that will matter next time: an exam that is over, what they plan to study next, how they like to learn. One fact per note, at most 200 characters. Never save passwords or private details. The learner can see and delete notes in Settings → Connection. |
| forgetA | Delete a saved note by its note_id (from "Where you left off" or from remember). Use it when a note is out of date or the learner asks. |
| save_ruleA | Save a rule about how godspeed's teacher teaches this learner, when they tell you a lasting preference ("use Python for code examples", "shorter answers"). It joins the learner's rules in Settings, for every lesson or for one space's lessons, and applies from the next turn. Not for one-off requests or facts about the learner (use remember for those). Don't ask first: the godspeed window shows it with Undo and Edit. |
| list_spacesA | List the learner's spaces, archived ones included, with file and lesson counts. |
| read_spaceA | Read one space: its description, how many files are in each status (pending, extracting, ready, failed, needs-whisper) and the files that could not be read, the topics of its knowledge graph with their fact counts, its deadlines (kind, date, time, done, and the files each covers), its lessons, and today's local date and time. A prep (a space made to get ready for one interview, assessment or test) also has its prep sheet: what the learner is preparing for, its day, what it is like, the topics with their ids and how each is going, what a mock is and the mocks taken. With files: true it lists every file instead (id, kind, status, size, path on disk). To find something inside the files, use search_space_files. |
| search_space_filesA | Search all of a space's files (readings, slides, transcripts, syllabus, assignments) by meaning and keywords. Returns the best passages, each with its id, file, place (page, slide, time or lines) and a short snippet. Open the ones you need with read_passage instead of reading whole files. |
| read_passageA | Read one passage of a space file by its id from search_space_files, or a file's passages at a place (file_id + range, e.g. a whole page). Returns the text with the ids of the passages before and after it, to keep reading. |
| create_spaceA | Create a space for one thing the learner is studying (a university course, a certification, a language). The description is shown to the teacher in every lesson of the space: level, the book or source they follow, its notation, what they are working toward (a course's exams, an interview). Add the space's deadlines with add_space_deadline. |
| update_spaceA | Rename a space, change its description, signature or icon, or archive / unarchive it (archiving hides it from the space switcher but keeps everything; an archived space takes no new lessons). |
| add_space_fileA | Copy a file from this computer into a space. Supported: .md .txt, code files (.c .py .java .js and the like), plain text with no extension, .pdf .docx .pptx .epub, .vtt .srt (transcripts) and .mp3 .m4a .wav .mp4 .mov (recordings, transcribed locally with whisper.cpp). Google Docs links (.gdoc) need saving as PDF or Word first. Study files are turned into facts by you, through wait_for_work. |
| set_space_file_kindA | Switch one or more space files between study material and space info. Study → info removes their facts; info → study builds them. A file that can't be switched is listed under refused; the others still switch. |
| add_space_deadlineA | Add one or more deadlines to a space (send all of a space's deadlines in one call): exams, homework, quizzes and projects, each with its date, its time if it has one, and which of the space's files (notes, readings) it covers. godspeed uses this to show the learner what is due and what to work on first. Take the deadlines, their dates and times from the space info files (a syllabus, a schedule), and never make up a date or time. Work out what a deadline covers from the space's own files first: a practice exam or review sheet, the schedule, announcements (search_space_files). Link the files they point to (file_ids), list files from before it that it doesn't cover in left_out_ids, and say in source which file it came from. Ask only when the files don't settle it. File ids come from read_space with files: true. A deadline that is refused is listed under refused with the reason; the others are still added, so send only the refused ones again. |
| update_space_deadlineA | Change a deadline's kind, name, date, time, covered files, left-out files, source or assignment file, or tick it done (done: true) or not done (done: false). file_ids and left_out_ids each replace the whole list (a file is never in both); time: null, source: null and assignment_file_id: null remove them. Work out coverage from the space's own files as add_space_deadline says; ask the learner when the files don't settle it. |
| remove_space_deadlineA | Remove one deadline from a space (its files and facts are not touched). |
| link_space_folderA | Link a folder on this computer to a space. Its supported files are added, and from then on new files dropped there are added automatically and edited files are rebuilt (checked every minute while godspeed is open). godspeed sorts them into study material and space info itself while reading them, and never changes the folder. Use set_space_file_kind only to correct a kind. |
| update_prepA | Change a prep's sheet when the learner asks (read_space shows it): move or clear its day, replace what it is like, change what a mock is, add topics, or remove topics by id. Only what you send changes. Refused while the sheet is being written. |
| start_lessonA | Start a new study session in godspeed taught by you: you teach what is new and give problems on what needs practice. The topic becomes the learner's first message. With space_id the lesson belongs to that space: it opens with the space's card and its facts go into that space's graph. In a prep, a topic named exactly like one on its sheet counts for that topic. You teach it through wait_for_work; the learner follows along in the godspeed window. |
| unlink_space_folderA | Unlink the space's folder: its files leave the space together with their facts. The folder itself is not touched. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 25 tools
Most tools target clearly distinct resources and actions, and the descriptions explicitly steer apart the riskiest pairs (remember vs save_rule, import_notes vs add_space_file, read_space vs search_space_files). A few clusters still require careful reading — the three 'save something' tools (remember, save_rule, import_notes) and the several read_* tools differ mainly by object, so a careless agent could pick wrong.
The dominant convention is a consistent verb_noun snake_case pattern (list_spaces, read_space, create_space, add_space_deadline, update_space_deadline, remove_space_deadline, link_space_folder). Three bare verbs (remember, forget, respond) deviate, but the overall pattern is predictable and readable.
At 25 tools the surface is at the heavy edge of the scale, spanning five sub-areas (work loop, spaces, files, deadlines, lessons, plus notes/rules). Each tool maps to a real operation, but the set is large enough that an agent must navigate several distinct workflows, pushing it past the comfortable 3-15 range.
Coverage is strong: full deadline CRUD, space create/read/update/list, file add/search/read/kind-switch/link-unlink, lesson list/read/start, graph search/import, and a complete work loop. Gaps are minor — no way to delete a space (only archive), no per-file removal, and save_rule has no list/delete counterpart — but these are workable around or handled in the app's Settings UI.