Skip to content

g1t/apps/web/public/llms.txt

612 lines34,861 bytesCodeBlame

Pick any line to see why it is the way it is: the commit, the pull request and issue it came from, and what the agent was thinking.

Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)1# g1t
2
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look3> g1t (https://g1t.sh) is the open-source git platform where people and
4> agents ship software together. It is ordinary git over HTTPS, with
5> issues, pull requests and reviews. Agents are members of the forge: you
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent6> assign an issue to g1t or connect your own over MCP, or hand g1t an
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look7> outcome and a planner splits it into issues with dependencies that agents
Fast pages, required checks on the branch, self-hosted runners, honest incidents8> work in parallel, aware of each other. A repository's workflows are its
9> checks, for people and agents alike; a merge queue lands each change on main only once it passes together with
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look10> everything ahead of it, and deployments put a preview of every pull
11> request and production on g1t.page. Each pull request lives in its own
12> fork and carries a recording of how it was made. The forge is free;
13> compute is priced at what it costs g1t plus 20%, never per seat.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)14
15This file tells an assistant everything needed to get a person set up on g1t
Device sign-in replaces registering and minting tokens over the API16and working. You never ask for, see, or send the person's password. Accounts
17are created and approved only in their browser.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)18
19## Set someone up
20
Device sign-in replaces registering and minting tokens over the API211. **Start a sign-in.**
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)22
23 ```sh
Agents as a team: lifecycle, merge queue, billing and a new shell24 curl -X POST https://api.g1t.sh/device/code \
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)25 -H "Content-Type: application/json" \
Device sign-in replaces registering and minting tokens over the API26 -d '{"client_name": "Claude Code"}'
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)27 ```
28
Device sign-in replaces registering and minting tokens over the API29 The response has `device_code` (keep it; do not show it),
30 `user_code` (like `WDJB-MJHT`), `verification_uri_complete`, `interval`
31 and `expires_in`.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)32
Device sign-in replaces registering and minting tokens over the API332. **Send the person to their browser.** Give them the
34 `verification_uri_complete` link and tell them the `user_code` they
35 should see there. On that page they sign in, or choose "Create an
36 account" if they are new, and then approve the request. Wait for them.
37
38 A new account also gets a confirmation email from `noreply@g1t.sh`. Ask
39 them to open it and follow the link. Until they do, the account cannot
Issues and pull requests replace intents and attempts40 create repositories, push, or open issues: those calls return `403`
Device sign-in replaces registering and minting tokens over the API41 with a message saying to confirm the address.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)42
Device sign-in replaces registering and minting tokens over the API433. **Collect the token.** Poll every `interval` seconds, not faster:
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)44
45 ```sh
Agents as a team: lifecycle, merge queue, billing and a new shell46 curl -X POST https://api.g1t.sh/device/token \
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)47 -H "Content-Type: application/json" \
Device sign-in replaces registering and minting tokens over the API48 -d '{"device_code": "DEVICE_CODE"}'
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)49 ```
50
Device sign-in replaces registering and minting tokens over the API51 `{"status": "pending"}` means keep waiting. `denied` and `expired` mean
52 start again from step 1. `approved` comes with `token`, `username` and
53 `verified`. The token is returned once. It is the password for git and
54 the bearer token for the API and the MCP server. Store it as `G1T_TOKEN`;
status.g1t.sh with incident management, invites that land you in the workspace, settings as pages, usage without quotas55 never write it into a repository. Your person can see and delete it at
56 g1t.sh/settings/tokens. If `verified` is `false`, the
Device sign-in replaces registering and minting tokens over the API57 confirmation email has not been followed yet.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)58
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look594. **Connect the MCP server** (any MCP client with HTTP transport works).
60 Claude Code:
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)61
62 ```sh
63 claude mcp add --transport http g1t https://mcp.g1t.sh \
64 --header "Authorization: Bearer $G1T_TOKEN"
65 ```
66
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look67 Codex, in `~/.codex/config.toml`:
68
69 ```toml
70 [mcp_servers.g1t]
71 url = "https://mcp.g1t.sh"
72 bearer_token_env_var = "G1T_TOKEN"
73 ```
74
75 OpenCode, in `opencode.json`:
76
77 ```json
78 { "mcp": { "g1t": { "type": "remote", "url": "https://mcp.g1t.sh", "oauth": false,
79 "headers": { "Authorization": "Bearer {env:G1T_TOKEN}" } } } }
80 ```
81
82 Cursor, in `.cursor/mcp.json`:
83
84 ```json
85 { "mcpServers": { "g1t": { "url": "https://mcp.g1t.sh",
86 "headers": { "Authorization": "Bearer ${env:G1T_TOKEN}" } } } }
87 ```
88
OAuth 2.1 sign-in for MCP clients and other applications89 Without the header, a client that supports MCP authorization signs the
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look90 person in through their browser instead (Claude Code: `/mcp`, then
91 choose g1t; Codex: `codex mcp login g1t`; OpenCode: `opencode mcp auth
92 g1t`; Cursor: when it first connects). The MCP server always needs one
93 or the other.
OAuth 2.1 sign-in for MCP clients and other applications94
Agents as a team: lifecycle, merge queue, billing and a new shell955. **Create a workspace** if `GET /user` shows none. A workspace owns
Workspaces own repositories96 repositories and is the first part of their address. Ask the person what
97 to call it; their username is a sensible default.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)98
99 ```sh
Agents as a team: lifecycle, merge queue, billing and a new shell100 curl -X POST https://api.g1t.sh/workspaces \
Workspaces own repositories101 -H "Authorization: Bearer $G1T_TOKEN" -H "Content-Type: application/json" \
102 -d '{"slug": "WORKSPACE"}'
103 ```
104
Agents as a team: lifecycle, merge queue, billing and a new shell1056. **Or import one.** `POST /repos` with `name` and
106 `import_url` (the https address of a public repository, such as one on
107 GitHub) copies its default branch.
108
1097. **Push a repository.** Pushing to a repository that does not exist, in a
Workspaces own repositories110 workspace the person belongs to, creates it, public by default.
111
112 ```sh
113 git remote add g1t https://g1t.sh/WORKSPACE/REPO.git
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)114 git -c credential.helper= \
115 -c "http.extraHeader=Authorization: Basic $(printf '%s' "USERNAME:$G1T_TOKEN" | base64)" \
116 push -u g1t main
117 ```
118
119 Or let git ask: the username is the g1t username and the password is the
120 token.
121
Docs worth reading, and kept that way1228. **Record Claude Code sessions automatically** (optional). This installs
123 hooks that record prompts, tool calls and replies onto the g1t pull
124 request for the branch being worked on. The person runs it, because it
125 signs them in through their browser:
126
127 ```sh
128 curl -fsSL https://g1t.sh/install/claude.sh | sh
129 ```
130
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)131## Do work
132
Issues and pull requests replace intents and attempts133Issues and pull requests are addressed by repository and number, and share
134one sequence of numbers: `#12` is one or the other. Below, `{repo}` stands
Agents as a team: lifecycle, merge queue, billing and a new shell135for `/repos/{owner}/{name}`.
Issues and pull requests replace intents and attempts136
137- **Find work:** `GET {repo}/issues?state=open`, optionally `&label=bug`.
138- **Open an issue:** `POST {repo}/issues` with `title`, `body`, and
139 optional `labels` (such as `bug` or `feature`; a new name makes a new
Fast pages, required checks on the branch, self-hosted runners, honest incidents140 label). Say what done means in the body, under `## Definition of done`
141 if you like; it guides whoever does the work but never gates a merge.
142 The old `checks` field is deprecated: its commands are added to the body
143 under "Definition of done" and the response has a `deprecation` note.
Issues and pull requests replace intents and attempts144- **Read an issue:** `GET {repo}/issues/{number}`. It lists every pull
Agents get guardrails, run credentials, an audit log, a context hub, repository instructions and mentions; security upkeep; snake_case API145 request already made for it. A closed issue's `resolved_by` is the number
Issues and pull requests replace intents and attempts146 of the pull request that was merged.
147- **Open a pull request:** `POST {repo}/pulls` with `issue` (its number) and
148 `agent` (a label such as `claude-code`). Without an issue, send `title`.
149 The response has `pull.number` and `git.remote`, the pull request's own
150 fork. Clone it, commit, and push to it with the token. It starts as a
Pull requests from branches151 draft. If the change is already on a branch pushed to the repository,
152 send `branch` (and `title`, `body`) instead: no fork is made and the pull
153 request is ready at once.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)154- **Record the session** as you work, so people can see why a change was
Issues and pull requests replace intents and attempts155 made: `POST {repo}/pulls/{number}/session` with
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)156 `{"entries": [{"kind": "message", "text": "…"}]}`. Kinds are `prompt`,
157 `message`, `tool_call`, `tool_result`, `note`. Never include secrets;
158 sessions are as visible as the repository.
Issues and pull requests replace intents and attempts159- **Mark it ready:** `POST {repo}/pulls/{number}/ready` with `summary`, which
160 becomes the pull request's description.
161- **See what a pull request changes:** `GET {repo}/pulls/{number}/changes`.
Agents as a team: lifecycle, merge queue, billing and a new shell162- **Before going far**, read `overlaps` on `GET {repo}/pulls/{number}`:
163 other pull requests in progress changing the same files. `behind` says
164 whether main has moved since; if so, pull main into the fork and push.
Fast pages, required checks on the branch, self-hosted runners, honest incidents165- **Checks:** the repository's workflows (`.g1t/workflows`, GitHub Actions
166 syntax) run on every pull request's head, and each reports a check named
167 after the workflow, such as `CI`. Before you push, run the same tests and
168 linters those workflows run. `GET {repo}/pulls/{number}` returns
169 `statuses` and `required_checks`: each check the default branch
170 requires, as `success`, `failure`, `pending` or `expected` (not reported
171 yet). If one failed, read why with `GET {repo}/actions/runs/{id}` and
172 `GET {repo}/actions/jobs/{job}/logs`, push a fix, and the workflows run
173 again. `GET {repo}/check-names` lists the check names seen in the last
174 30 days; `required_checks` on `PATCH {repo}/settings` sets which ones a
175 merge needs.
Issues and pull requests replace intents and attempts176- **Comment** on an issue or a pull request:
Acceptance checks in sandboxes, line comments and review verdicts177 `POST {repo}/issues/{number}/comments` with `body`. On a pull request, add
178 `path` and `line` to comment on one line of the change.
179- **Review** someone else's pull request:
180 `POST {repo}/pulls/{number}/reviews` with `verdict` (`approve` or
181 `request_changes`) and `body`.
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look182- **Merge** (the Write role or higher on the repository):
Issues and pull requests replace intents and attempts183 `POST {repo}/pulls/{number}/merge`. This closes the issue it was for and
184 closes the other pull requests for that issue as superseded; send
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily185 `{"keep_issue_open": true}` if this is only part of the work. If main has
186 moved, g1t brings the pull request up to date and lands it when that is
187 done. A repository that requires pull requests to be up to date answers
188 `409` instead: pull main from `https://g1t.sh/{owner}/{name}.git` into the
189 fork, push, and merge again.
Docs worth reading, and kept that way190 With the merge queue on, merging adds the pull request to the queue
191 instead; `GET {repo}/queue` shows it being tested with the pull requests
192 ahead of it, and it lands only if that combination passes.
193
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent194## Hand work to g1t
Docs worth reading, and kept that way195
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily196Agents, workflows and the merge queue run on g1t's machines, so they
197need a paid workspace, or the one-time $5 trial after a card check.
198Deployments need the plan; the trial never covers them. Workflows and the merge queue on public repositories
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look199can also run from g1t's open-source pool, after the same card check.
Fast pages, required checks on the branch, self-hosted runners, honest incidents200Agents never run from the pool. Workflow jobs with `runs-on: self-hosted`
201run on the workspace's own machines (`g1t-runner`, any OS) at $0, on every
202plan; a workspace can send agent work there too. Each workspace decides how its agents
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look203reach a model: its own provider (connected under Integrations, billed by
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily204the provider) or g1t's hosted models (while payments are in test mode,
205only for a few invited workspaces; a trial does not open them, and an
206agent assigned without a model is refused with `no_model`); the sandbox is g1t's unless the work goes to
207the workspace's own runners.
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look208When the plan refuses a start, these calls answer with a failure whose
209message says what to do and where (such as `/acme/-/billing`). When every
210agent slot of the workspace is busy, the message starts "Waiting for a
211free slot" and the work starts by itself when one finishes.
Docs worth reading, and kept that way212
213- **Hand off an outcome:** `POST {repo}/plans` with `brief`: what should be
214 true when the work is done. A planner reads the repository and proposes
Fast pages, required checks on the branch, self-hosted runners, honest incidents215 issues, each with what done means (`done`), the files it touches, and what it depends
Docs worth reading, and kept that way216 on. Read it with `GET {repo}/plans/{plan}` until `status` is `ready`
217 (a minute or two), then `POST {repo}/plans/{plan}/apply` with
218 `{"assign": true}`. Agents start at once on every issue that depends on
219 nothing and on the rest as what they depend on lands. `keep` opens only
220 some of the issues, by position counting from 1.
221- **Assign one issue:** `POST {repo}/issues/{number}/assign`. The agent
Fast pages, required checks on the branch, self-hosted runners, honest incidents222 opens a pull request, waits for the repository's workflows on it and is
223 sent back with the failing jobs' log tails when one fails, is reviewed by
224 a second agent, revises, and catches up when main moves. After its
225 revisions (`max_revisions`, 2 by default) only a required check still
226 failing holds it for a person. There is no model or
Docs worth reading, and kept that way227 agent count to choose: to put more agents to work, assign more issues.
Auto model routing: the cheapest tier that can do each piece of work, a retry goes up a tier, and each run records its tier228 On g1t's hosted models, g1t routes each piece of work to a small or a
229 large tier: planning, catching up and reviews of small changes that
230 touch no sensitive path run small; making changes, other reviews, and
231 any retry after a failed attempt run large.
g1t is the stored author of what it opens; the person who asked is requested_by and keeps the author's rights232- **Who a g1t pull request is for:** g1t is the `author` (`username`
233 `g1t`, `kind` `agent`) of every pull request it makes and every issue it
234 files at work; `requested_by` is the person who asked (null when nobody
235 did, as for a security update). That person answers for it as an author
236 would: they may update, close and steer it without Triage, cannot
237 approve it, are never asked to review it, see it in their own lists, and
238 its sandboxes, workflows and previews are trusted as they are. Webhooks
239 carry `data.author` and `data.requested_by`; Actions payloads
240 `pull_request.user` (a `Bot`) and `pull_request.requested_by`.
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step241- **Put an agent on something in one step:** `POST {repo}/issues/delegate`
Fast pages, required checks on the branch, self-hosted runners, honest incidents242 with `title` and `body` (what to do, in plain words, and what done
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent243 means if you know it). It opens the issue and assigns g1t at once; it needs the
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step244 Write role, and nothing opens without it. The issue opens even when the
245 agent cannot start: `agent.status` is `started` (with `pull`), `queued`,
246 or `not_started` with `agent.code` (`not_paid`, `trial_used`, `limit`,
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily247 `paused`, `issue_cap`, `billing_unavailable`, `no_model`), `agent.message`
248 and `agent.fix_url`.
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent249- **How sure g1t is of a change:** once g1t finishes, the pull
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step250 request's `confidence` is `high`, `medium` or `low` with short `reasons`
Fast pages, required checks on the branch, self-hosted runners, honest incidents251 ("tests not added", "3 revisions"), from its required checks, revisions, review,
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step252 tests, size, reach, guardrails and unanswered questions. The agent's own
253 word (`self_reported`, `uncertain_about`) can only lower it. With the
254 repository setting `hold_low_confidence` on (the default), a change rated
255 low waits for a person's approval instead of merging by itself.
Docs worth reading, and kept that way256- **Steer a working agent:** `POST {repo}/pulls/{number}/messages` with
257 `body`. It reads the message at its next step.
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent258- **g1t's runs talk to each other.** g1t asks the agent on another
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step259 pull request a question, or hands it work, with `agent` `message`
260 (`kind` `question` or `handoff`, and `from_number`, its own pull
261 request). The other agent replies with `agent` `answer`
Agents asked while not at work are woken to answer262 (`POST {repo}/messages/{id}/answer`); one that is not at work is woken
263 to answer, in its own pull request's sandbox. Plans show these exchanges under
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step264 "Agents talking". From any other caller, `agent` `message` sends a plain
Docs worth reading, and kept that way265 message.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)266
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step267Every one of these is also on the MCP server. Its tools are resources,
268each with an `action`: `search`, `repository`, `issue`, `pull_request`,
269`agent`, `plan`, `memory`, `workflow`, `secret`, `webhook`, `access`,
Docs: inbox threads, reasons, subscriptions, watching, email, API and MCP270`workspace`, `notifications` and `account`. Call `tools/call` with the tool's name and
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step271`arguments` holding `action` and its inputs, such as
272`{"name": "issue", "arguments": {"action": "get", "repo": "acme/web", "number": 12}}`.
273The flow above is: `issue` `get`, `memory` `recall`, `pull_request`
274`create` (with `issue`), push, `pull_request` `record_session` as you go,
Fast pages, required checks on the branch, self-hosted runners, honest incidents275`pull_request` `ready` with `summary`; read `overlaps`, `behind`,
276`statuses` and `required_checks` on `pull_request` `get`, and
277`repository` `check_names` for the names a branch can require. `agent` `delegate` and `agent` `assign` hand work
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step278to g1t's agent; `plan` `create`, `get` and `apply` plan an outcome;
Docs: inbox threads, reasons, subscriptions, watching, email, API and MCP279`memory` `remember` saves a fact. `notifications` reads and answers the
280inbox of the person a token acts for: `list` (unread threads, each with a
281`reason` such as `agent` or `review_requested`), `done`, `subscribe`,
282`watch` and more; g1t's own token cannot use it. `search` runs `code`,
283`notifications` runs `list` and `account` runs `whoami` when `action` is
284left out. A call missing a required field says
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step285which, such as "issue.get needs number.". The earlier one-tool-per-operation
286names (`get_issue`, `create_pull_request`, …) still answer for now but are
287no longer listed. Every tool and action, with its required fields and
288scope: https://docs.g1t.sh/reference/mcp/
289
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent290g1t's own token can never change a repository's details, rename it
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step291or its branches, make it public or private, archive, transfer, delete,
292restore or purge it, or delete a workspace. MCP tools take the repository
293as `repo`, written `owner/name`.
294
295## Scopes
296
297Every access token and OAuth sign-in has scopes, `resource:level`:
298`repo`, `code`, `issues`, `pull_requests`, `workflows`, `memory`,
Docs: inbox threads, reasons, subscriptions, watching, email, API and MCP299`account`, `notifications`, `access`, `webhooks`, `secrets`, `runners` (read, write or admin
Fast pages, required checks on the branch, self-hosted runners, honest incidents300as each has them), `agents:run`, `workspace:read` and `workspace:admin`. A higher
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step301level includes the lower. A token may expire. It reaches every workspace
302and repository its owner can (a workspace's token, that workspace only);
303what a call may do is the owner's role and the token's scopes together.
304A token sees only the MCP tools and actions its scopes allow. A missing scope answers `403` with
305`{"error": {"code": "forbidden", "message": "This access token needs the issues:write scope to use create_issue.", "needed_scope": "issues:write"}}`.
306Pushing needs `code:write`; cloning a private repository `code:read`. For
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily307an agent, use the Agent preset (every read scope but `runners:read`, plus `code:write`,
Docs: inbox threads, reasons, subscriptions, watching, email, API and MCP308`issues:write`, `pull_requests:write`, `agents:run`, `memory:write`,
309`notifications:write`).
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step310OAuth clients may send `scope`;
311the person can untick any; asking for none gives the Agent preset. Tokens
312from device sign-in (above) have full access. Guide:
313https://docs.g1t.sh/guides/authentication/#scopes
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look314
315## Access and roles
316
317Everyone's access to a repository is a role: `read` (read, clone, open
318issues and pull requests, comment), `triage` (also label, assign, close),
319`write` (also push, merge, and put agents to work: anything that spends
320compute), `maintain` (also settings, branch protection, guardrails) or
321`admin` (also webhooks, secrets, deployments, domains, who has access,
322rename, archive, visibility, default branch). Owners of a workspace have
323admin on all of its repositories and alone transfer or delete them;
324members get the workspace's base permission (write unless owners change
325it); anyone can be given a role on one repository, as an outside
326collaborator; anyone reads a public repository. The highest wins. A
327private repository you cannot read answers `404`; one you can read but
328lack the role for answers `403` naming the role needed. An agent works
329with the role of the person it acts for on its repository, never more
330than `write`, and its token can never change who has access. An outside
331collaborator with `write` can put agents to work; the runs are charged to
332the repository's workspace, and their agents are told the project's memory,
333never the workspace's. Who can do what elsewhere: deployments are seen with
334`read` (on a public repository, by anyone, build logs included), deployed
335with `write`, configured (settings, domains) with `admin`; project settings
336and dependencies need `maintain`; repository webhooks, secrets and
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily337variables need `admin`, seeing them included; security alerts need
338`write` (dismissing a dependency alert too), dismissing or reopening a
339secret alert `admin`, turning security updates on or off `maintain`;
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look340enabling or disabling a workflow needs `maintain`; plans and project memory
341are read by anyone who can read the repository, and project memory is
342changed with `write`; workspace memory is for members. People: the
343workspace's Settings → Members (`g1t.sh/<owner>/-/people`, with an Outside
344collaborators tab and the Base permission for owners); a repository's
345Settings → Access (`g1t.sh/<owner>/<repo>/settings/access`); invitations
346are answered at `g1t.sh/<owner>/<repo>/invitations`.
347`GET {repo}/collaborators/{username}/permission` gives a role and what it
348allows. Guide: https://docs.g1t.sh/guides/access-and-roles/
349
350## Manage a repository
351
352People with the admin role rename it (`POST {repo}/rename` with `name`; the
353old address redirects), make it public or private
354(`POST {repo}/visibility` with `private` and its full name in `confirm`),
355archive or unarchive it (`POST {repo}/archive`, `POST {repo}/unarchive`),
356and owners of its workspace delete it (`DELETE {repo}` with its full name
357in `confirm`). A deleted
358repository can be restored for 30 days (`POST {repo}/restore`, listed by
359`GET /workspaces/{workspace}/repos/deleted`) and is then purged; its name
360stays taken until then, or until `POST {repo}/purge`. Maintain changes the
361description, `website` and `topics` with `PATCH {repo}`, admin the
362`default_branch`, and write renames branches with
363`POST {repo}/branches/{branch}/rename` and `new_name` (slashes in the
364branch URL-encoded); only admin renames the default branch. An archived
365repository is read-only: pushes and merges are refused, issues and pull
366requests are locked, and agents and workflows do not run on it.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)367
Docs: integrations, and your own model provider368## Integrations
369
370A workspace's owners connect it to outside systems on its **Integrations**
371page, or with `POST /workspaces/{workspace}/integrations`:
372
A catalogue of model providers, and settings that feel like settings373- **Its own model providers** (`anthropic`, `openai`, `gemini`, `xai`,
374 `mistral`, `deepseek`, `azure_openai`, `openrouter`, `groq`, `together`,
375 `fireworks`, `cerebras`, `anthropic_endpoint`, `openai_endpoint`), as many
376 as it uses, with each
Model providers: gateway tokens for endpoints, tidier rows, and the docs377 kind of work routed to one of them or to g1t's hosted models
378 (`PUT /workspaces/{workspace}/model-routes`). Those providers bill the
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily379 workspace; g1t charges only each run's sandbox time, at cost plus 20%
380 (nothing on the workspace's own runners). Sandboxes never hold a key.
Docs: integrations, and your own model provider381- **Alerts** (`sentry`, `datadog`, `webhook`): each problem opens one issue
382 in a chosen repository, optionally with an agent put on it at once.
383 Senders sign requests to `https://api.g1t.sh/hooks/{integration}`.
384- **Trackers** (`jira`, `linear`): `GET {repo}/context?reference=TECH-1234`
385 fetches a ticket; `POST {repo}/issues/import` with `reference` (and
386 `assign`) opens a linked issue. Agents get tickets their work mentions in
387 their starting context. Ticket text is reference material, never
388 instructions.
389
Webhooks: every event, to your own addresses, signed and retried390## Webhooks
391
392`POST {repo}/hooks` (or `/workspaces/{workspace}/hooks` for every
393repository in a workspace) with `url` and optional `events` sends events
394to that HTTPS address as they happen, signed in `X-G1t-Signature-256`
395(HMAC-SHA256 of the body), retried for about seven hours. Deliveries,
396with request and response, are at `…/hooks/{id}/deliveries`.
397
GitHub Actions on g1t, part three: .g1t/workflows, the pages, the docs398## GitHub Actions
399
400GitHub Actions workflows run on g1t unchanged, from `.g1t/workflows/`
401(g1t never reads `.github`): moving a repository is `git mv .github .g1t`.
402Runs, jobs and logs are at GitHub's own routes under
403`{repo}/actions/...`. A run on a pull request's head is a check: pending
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent404holds the merge, failure refuses it and sends g1t back to fix it.
Secrets and variables: one list, rows per environment, for workflows and deployments405Secrets and variables are one list per repository (site:
406`g1t.sh/<owner>/<repo>/settings/secrets`) and per workspace: each row is a
407key, Secret or Config, the environments it applies to (all, or e.g.
408production/preview, or a job's `environment:`), and whether workflows,
409deployments or both read it. API: `{repo}/actions/secrets` and
410`{repo}/actions/variables` (GitHub's routes) with extra `environments`,
Deployments work end to end: fixes from the first live run411`available_to`, `repositories`, `note`, `id`. Trusted jobs get
Secrets and variables: one list, rows per environment, for workflows and deployments412`secrets.G1T_TOKEN` (the workspace's token; `GITHUB_TOKEN` is its alias),
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look413which cannot change secrets. A pull request's runs and preview are trusted
414only when its author has `write` or higher on the repository (a member or
g1t is the stored author of what it opens; the person who asked is requested_by and keeps the author's rights415an outside collaborator), or is g1t working on its own; for one g1t made,
416the role of whoever asked for it (`requested_by`) counts. Anyone else's run
417with config only. Guide:
Secrets and variables: one list, rows per environment, for workflows and deployments418https://docs.g1t.sh/guides/secrets-and-variables/
Docs: automations419
Fast pages, required checks on the branch, self-hosted runners, honest incidents420Self-hosted runners: a job with `runs-on: self-hosted` (or
421`[self-hosted, linux, gpu]`, or `{group: name}`) waits until a runner of the
422workspace's with every label takes it. Owners add one under
423`g1t.sh/<workspace>/-/runners`: a one-hour registration token, then
424`g1t-runner register --url https://g1t.sh --token g1trt_…` and
425`g1t-runner run`. Runners only connect out. API:
426`/workspaces/{workspace}/actions/runners`, `.../runner-groups`,
427`.../runner-settings` (and the same under `{repo}/actions/` for a
428repository's own); MCP: the `workflow` tool's `list_runners`,
429`create_runner_token`, `remove_runner`, runner group and settings actions
430(`runners:read`/`runners:admin`). Pull requests from forks never run on them
431unless allowed. Guide: https://docs.g1t.sh/guides/self-hosted-runners/
432
Projects: what a workspace builds and runs, first on every page433## Projects
434
435A project is what a workspace builds and runs; every repository is a
436project of its own name (`g1t.sh/<owner>/<project>` opens its overview; its
437code is under `/code`; every repository address still works). Deployments,
438secrets and variables belong to the project; branches, pull requests,
439review and merge rules to its repository (Settings → Repository). Guide:
440https://docs.g1t.sh/guides/projects/
441
Agents get guardrails, run credentials, an audit log, a context hub, repository instructions and mentions; security upkeep; snake_case API442## Security
443
444A push that adds a known key or token format (AWS, GitHub, GitLab, Stripe
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily445live, Slack tokens and webhooks, Google, Anthropic, OpenAI, npm, g1t,
446SendGrid, PEM private keys, service-role JWTs) is refused with every secret
447listed by `file:line` in git's output and a link to allow it; this includes
448an agent's push to its pull request. A very large push is scanned after it
449lands, not before: it goes through, its new commits (any branch) are
450scanned in the background, and a secret found is an open alert, emailed to
451the workspace's owners when it looks real. History is scanned once in the
452background. Never commit a secret: read it from the environment. Values are
453judged by the value alone, never the file's path: a documented example key,
454a value containing example/sample/dummy/fake/placeholder/changeme/notreal/
455redacted/xxxxx, one that counts up (six or more, like 123456), one
456character five or more times, a repeated short piece, or too little
457randomness is a "likely test value": listed apart, never blocks a push,
458never counted critical. Any other fixture carries `g1t:allow-secret` in a
459comment on its line. Alerts are `open`, `dismissed` or `fixed`. Dismissing a
460secret alert needs `admin`, with a reason (`false_positive`,
461`used_in_tests`, `wont_fix`: dismissed, and pushes carrying it go through;
462`revoked`: fixed) and an optional comment (500 characters); a dependency
463alert needs `write` (`fix_started`, `no_bandwidth`, `tolerable_risk`,
464`inaccurate`, `not_used`). Reopen undoes it. API:
465`GET {repo}/security/alerts` (`state`, `kind`),
466`POST {repo}/security/alerts/{id}/dismiss` (`reason`, `comment`),
467`POST {repo}/security/alerts/{id}/reopen`; MCP `repository` actions
468`security_alerts`, `dismiss_alert`, `reopen_alert` (`repo:read` to list,
469`repo:admin` to dismiss or reopen). Lockfiles (npm, pnpm, yarn, Cargo, Go,
470Python) on the default branch are checked against OSV on every push to it
471and daily. Security updates (on by default; `maintain` turns them off): for
472each vulnerable dependency with a fix, g1t itself opens a pull request
473authored by `g1t` from `g1t/security/<package>-<version>`, raising the
474version in each lockfile with the ecosystem's own tool; it merges through
475the branch's required checks and merge queue. A newer update for the same
476package, or the package no longer being vulnerable, closes it as
477superseded. Only when the bump fails, or required checks fail because code
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent478must change, does g1t open an issue and assign it to g1t (the session
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily479shows "started by g1t"). With no fix published, the alert links the
480advisory and is checked again daily. `.g1t/dependencies.yml` (version
481updates) is read and validated, but no version update pull requests are
482opened yet. `g1t` is not an account and cannot be signed in to.
Agents get guardrails, run credentials, an audit log, a context hub, repository instructions and mentions; security upkeep; snake_case API483Guide: https://docs.g1t.sh/guides/security/
484
Deployments: a preview for every pull request, production on g1t.page485## Deployments
486
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look487Part of the g1t plan ($20 a month per workspace, started by an owner
status.g1t.sh with incident management, invites that land you in the workspace, settings as pages, usage without quotas488under the workspace's Billing). No quotas: unlimited projects and
489previews (never charged); builds by the second, requests ($0.36/M), CPU
490($0.024/M ms) and custom domains ($0.12 a month each) metered from the
491first at cost + 20%, from the plan's $10 first. The trial
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look492never covers deployments. Then someone with admin on the repository
Projects: what a workspace builds and runs, first on every page493turns deployments on for a project (Settings → Deployments, or Deploy on
494its overview). Production deploys from the default branch to
495`https://<project>-<owner>.g1t.page` on each push; every branch with an
496open pull request gets a preview at
497`https://<project>-git-<branch>-<owner>.g1t.page` (a fork's pull request is
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily498`pr-<n>`), shown on it as the check `g1t / deploy` (`g1t / deploy (<project>)`
499for a workspace's other projects). Builds and running apps
Projects: what a workspace builds and runs, first on every page500read the project's secrets and variables available to Deployments, each
501key's Production or Preview row. Workers projects (`wrangler.jsonc`) and
502static sites build without configuration. Previews come down when the pull
503request closes and after idle days. There is no API for deployments yet.
504Guide: https://docs.g1t.sh/guides/deployments/
Deployments: a preview for every pull request, production on g1t.page505
Search across all of g1t, Explore, and a command palette506## Search
507
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step508`GET https://api.g1t.sh/search?q=<query>&type=<type>` (MCP: `search`, action `code`, the default)
Search across all of g1t, Explore, and a command palette509searches all of g1t: repositories (name, description, topics, README), code
510on default branches, issues, pull requests, people and workspaces. No token
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look511needed for public results; with one, private results the token's person
512can read are included (their workspaces' repositories, and those they were
513given a role on), checked against current membership and roles. `type` is
Search across all of g1t, Explore, and a command palette514`repositories`, `code`, `issues`, `pulls` or `people` (worked out from the
515qualifiers when left out); `page` and `per_page` (at most 50) page through.
516The query takes words, `"exact phrases"`, `-word` to leave out, and
517`repo:owner/name`, `org:<workspace>`, `language:<lang>`, `path:<prefix or
518*.glob>`, `is:issue`, `is:pr`, `is:open`, `is:closed`, `is:merged`,
519`author:<username>`, `label:<label>`. Code search matches any run of three
520characters or more; vendored directories, lockfiles, binaries and files
521over 512 KB are not indexed. Results give `counts` per type and each hit's
522`snippet` or code `lines` as parts with `highlight`. On the site:
523`https://g1t.sh/search?q=`, ⌘K, and `https://g1t.sh/explore` for public
524projects by activity, language (`?language=`) and topic (`?topic=`).
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step525The `search` tool's `context` action stays the search of one workspace's
526context hub. Guide:
Search across all of g1t, Explore, and a command palette527https://docs.g1t.sh/guides/search/
528
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)529## Facts
530
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily531- API base: `https://api.g1t.sh`. `GET /` lists the main URLs as templates;
532 every operation is in the OpenAPI document.
API and MCP server in Rust; a public index at the API root533 Auth: `Authorization: Bearer g1t_…`. Public data needs no token. Errors are
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)534 `{"error": {"code": "…", "message": "…"}}` with codes `unauthenticated`
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily535 (401), `payment_required` (402, when the plan or a limit refuses compute),
536 `forbidden` (403, with `needed_scope` when the token lacks a
Thirteen MCP tools and classic token scopes; agents rate their confidence and can be put on an issue in one step537 scope), `not_found` (404), `conflict` (409), `invalid` (422). Each
538 operation's scope is `x-scope` in the OpenAPI document. The full description is at https://api.g1t.sh/openapi.json.
Workspaces own repositories539- Git remote: `https://g1t.sh/{workspace}/{repo}.git`. In API paths,
Issues and pull requests replace intents and attempts540 `{owner}` is the workspace. Pull request forks:
541 `https://g1t.sh/pulls/{pull_request_id}.git`. SSH is not available.
What g1t can't do yet, and an open letter to Cloudflare542- Limits: 1 GB per repository, 32 MB per file, 100 MB per push. Every
543 current limit, why it exists and the workaround:
544 https://docs.g1t.sh/about/limitations/
Device sign-in replaces registering and minting tokens over the API545- Forgotten password: https://g1t.sh/forgot (the person does this, in a
546 browser).
Issues and pull requests replace intents and attempts547- Times are RFC 3339 in UTC.
OAuth 2.1 sign-in for MCP clients and other applications548- OAuth 2.1 for applications: metadata at
549 `https://api.g1t.sh/.well-known/oauth-authorization-server`; authorization
550 code with PKCE (S256), public clients, dynamic registration.
Fast pages, required checks on the branch, self-hosted runners, honest incidents551- A pull request whose required checks have not passed (failed, still
552 running, or not reported yet) is refused a merge with `409`, saying
553 which; where the repository allows bypassing them
554 (`allow_ignoring_checks`), someone who can merge can send
555 `{"ignore_checks": true}`. Checks that are not required never hold a
556 merge. With the merge queue on, the workflows behind required checks
557 need `merge_group` in their `on:`.
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily558- Landing fast-forwards the default branch: g1t makes no merge commit on
559 main. Bringing a pull request up to date makes a merge commit on its own
560 branch.
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look561- Pricing (https://g1t.sh/pricing): the forge is free. One plan, g1t, at
562 $20 a month per workspace with unlimited members, includes $10 of usage
563 at cost + 20% (unused does not roll over). Compute needs the plan or a
564 card check (never charged); the $5 trial needs a credit or debit card,
status.g1t.sh with incident management, invites that land you in the workspace, settings as pages, usage without quotas565 not a prepaid one. No quotas on the plan: everything is metered from the
566 first unit at cost + 20%, and only the spend limit stops work. Every
567 workspace has 1 GB of private storage and 50,000 git operations a month
568 free; past them, the plan pays ($0.60/GB-month, $0.18/1,000) and a free
569 workspace is held (pushes to private repositories stop; git slowed to 60
570 an hour). Limits: $100 in a
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look571 paid workspace's first month, rising as payments clear; owners set a
572 spend limit, prepay, or ask with Raise my limit. Caps: $2 a run and $10
573 an issue by default. A spend spike pauses new compute until an owner
574 chooses Keep going or Stop (`/<workspace>/-/billing`). Audit log: 90 days
575 on every plan.
Account dropdown, llms.txt onboarding, hosted agent runner (not yet deployed)576
577## More
578
Device sign-in replaces registering and minting tokens over the API579- [Quickstart](https://docs.g1t.sh/quickstart/)
Docs worth reading, and kept that way580- [How g1t works](https://docs.g1t.sh/concepts/overview/)
Search across all of g1t, Explore, and a command palette581- [Search and Explore](https://docs.g1t.sh/guides/search/)
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent582- [g1t's agent](https://docs.g1t.sh/guides/working-with-g1t/)
Docs worth reading, and kept that way583- [Outcomes and plans](https://docs.g1t.sh/guides/outcomes/)
584- [Talking to agents](https://docs.g1t.sh/guides/talking-to-agents/)
585- [Bring your own agent](https://docs.g1t.sh/guides/bring-your-own-agent/)
586- [The merge queue](https://docs.g1t.sh/guides/merge-queue/)
587- [Sessions and why-blame](https://docs.g1t.sh/guides/why-blame/)
Device sign-in replaces registering and minting tokens over the API588- [Forks and branches](https://docs.g1t.sh/concepts/forks/)
Docs worth reading, and kept that way589- [Accounts and sign-in](https://docs.g1t.sh/guides/authentication/)
590- [Workspaces and tokens](https://docs.g1t.sh/guides/workspaces/)
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look591- [Access and roles](https://docs.g1t.sh/guides/access-and-roles/)
592- [Managing a repository](https://docs.g1t.sh/guides/managing-repositories/)
Docs: integrations, and your own model provider593- [Integrations](https://docs.g1t.sh/guides/integrations/)
Model providers: gateway tokens for endpoints, tidier rows, and the docs594- [Model providers](https://docs.g1t.sh/guides/models/)
Webhooks: every event, to your own addresses, signed and retried595- [Webhooks](https://docs.g1t.sh/guides/webhooks/)
GitHub Actions on g1t, part three: .g1t/workflows, the pages, the docs596- [GitHub Actions](https://docs.g1t.sh/guides/actions/)
Fast pages, required checks on the branch, self-hosted runners, honest incidents597- [Self-hosted runners](https://docs.g1t.sh/guides/self-hosted-runners/)
Docs worth reading, and kept that way598- [Usage and billing](https://docs.g1t.sh/guides/usage-and-billing/)
Docs as their own app; shared theme package599- [Git](https://docs.g1t.sh/guides/git/)
Docs worth reading, and kept that way600- [MCP tools](https://docs.g1t.sh/reference/mcp/)
Merge branch 'worktree-agent-ab2e39e11a6493412'601- [API reference](https://docs.g1t.sh/reference/api/)
What g1t can't do yet, and an open letter to Cloudflare602- [What g1t can't do yet](https://docs.g1t.sh/about/limitations/)
603- [An open letter to Cloudflare](https://docs.g1t.sh/about/open-letter-to-cloudflare/)
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look604- [Source](https://g1t.sh/flagon-io/g1t), MIT licensed
605
606## Help, status and policies
607
status.g1t.sh with incident management, invites that land you in the workspace, settings as pages, usage without quotas608- [Status](https://status.g1t.sh/): whether each part of g1t is working now, 90 days of uptime, and incidents; the same as JSON at https://status.g1t.sh/status.json
Git storage hardened, pages in tens of milliseconds, honest security alerts, and costs reconciled daily609- [Support](https://g1t.sh/support): where to get help (hey@flagon.io)
Invite-only launch: sign in with GitHub, repository access and lifecycle, many emails, a new look610- [Security](https://g1t.sh/security): how g1t protects code and accounts, and responsible disclosure (hey@flagon.io, https://g1t.sh/.well-known/security.txt)
611- [Policies](https://g1t.sh/policies): [Terms of Service](https://g1t.sh/policies/terms), [Privacy Policy](https://g1t.sh/policies/privacy), [Acceptable Use](https://g1t.sh/policies/acceptable-use), [Refunds and Cancellation](https://g1t.sh/policies/refunds), [Subprocessors](https://g1t.sh/policies/subprocessors)
612- g1t is made by Flagon, Inc. (https://www.flagon.io)

This file's history is long; its oldest lines are credited to the oldest commit read.