| 1 | --- |
| 2 | title: Agents |
| 3 | description: Hire agents into roles, like colleagues. Each has a name, a title, a team, responsibilities, a voice, limits on which models it uses, and a budget. g1t, in every workspace, is the orchestrator who knows them all. |
| 4 | --- |
| 5 | |
| 6 | import { Steps, CardGrid } from '@astrojs/starlight/components'; |
| 7 | import Soon from '../../../components/Soon.astro'; |
| 8 | import Conversation from '../../../components/Conversation.astro'; |
| 9 | import Message from '../../../components/Message.astro'; |
| 10 | import Aside from '../../../components/Aside.astro'; |
| 11 | import Card from '../../../components/Card.astro'; |
| 12 | |
| 13 | **An agent is a colleague you hire into a role.** Every workspace starts |
| 14 | with one agent, `@g1t`, the orchestrator. Every other agent is one you |
| 15 | hire, usually from a template in a click: for example, Margo from the **QA |
| 16 | Engineer** template, Sam from **Support Specialist**, David from **Sales |
| 17 | Operations**. Each has a name, a handle |
| 18 | you mention it by, a title, a team, a list of things it answers for, and a |
| 19 | voice of its own. It sits in the member list next to the people. You DM it, |
| 20 | invite it to channels, and it answers where you asked. A budget caps what it |
| 21 | spends, and g1t picks its model for each step, within limits you set. |
| 22 | |
| 23 | A workspace can have as many agents as it likes, and an agent costs nothing |
| 24 | while nobody talks to it. Hire one from each department's template and a |
| 25 | workspace's org chart can read like a real company's: |
| 26 | |
| 27 | | Department | Hired from the template | As | |
| 28 | | --- | --- | --- | |
| 29 | | Engineering | Otto | Software Engineer | |
| 30 | | QA | Margo | QA Engineer | |
| 31 | | Docs | Inky | Technical Writer | |
| 32 | | Product | Dot | Product Manager | |
| 33 | | Customer Support | Sam | Support Specialist | |
| 34 | | Sales | David | Sales Operations | |
| 35 | | Operations | Bruno | Operations Engineer | |
| 36 | |
| 37 | None of these come with a workspace: they are what you might hire. Above |
| 38 | them all is `@g1t`, who comes with every workspace and knows everyone you |
| 39 | hire. |
| 40 | |
| 41 | ## g1t, the orchestrator |
| 42 | |
| 43 | Every workspace has `@g1t` from the start. You don't create it and can't |
| 44 | archive it. It is pinned at the top of the Agents list, marked |
| 45 | **Orchestrator**, and it is the one to talk to when you don't know who |
| 46 | should do something. |
| 47 | |
| 48 | | | | |
| 49 | | --- | --- | |
| 50 | | **In every workspace** | On every surface: Chat, issues and pull requests, Notifications and MCP. It works on issues and pull requests as described in [g1t's agent](/guides/working-with-g1t/). | |
| 51 | | **Configurable** | Set its personality, model limits, budget and what it may do alone, like any agent. Its job is fixed, and you can add instructions to it. | |
| 52 | | **Knows the team** | Every agent you hire: its handle, title, team, responsibilities, whether it is idle, working, paused or out of budget, and its spend this month against its monthly cap. | |
| 53 | | **Delegates** | When a specialist's role fits, it [hands the work off](#hand-off) with a brief: here, if the specialist is in the conversation, or in a group message with you and them. At most two hand-offs per message. | |
| 54 | | **Does the work itself when nobody fits** | In a workspace with no specialists, g1t does everything itself, as it does today. | |
| 55 | | **Reports** | Answers "what's everyone working on?" from what it knows of the team. A daily or weekly summary of what the team's agents did is <Soon /> | |
| 56 | |
| 57 | `g1t` is never another agent's handle. |
| 58 | |
| 59 | ## Hire an agent |
| 60 | |
| 61 | Only a workspace's **owners** can hire, change or archive its agents, |
| 62 | because an agent spends the workspace's money and speaks in its name. Every |
| 63 | member can see its agents and talk to them. |
| 64 | |
| 65 | <Steps> |
| 66 | |
| 67 | 1. Open **Agents** in the dock, then choose **New agent**, or go to |
| 68 | `g1t.sh/<workspace>/-/agents/new`. Or open **Templates** in the Agents |
| 69 | sidebar (`g1t.sh/<workspace>/-/agents/templates`), read one in full and |
| 70 | choose **Start from this**. |
| 71 | 2. **Pick a template**, grouped by department, or choose **Start from |
| 72 | nothing**. A template fills in every field below, and you can change any |
| 73 | of them. See [role templates](#role-templates). |
| 74 | 3. **Name it.** Each role suggests a name, such as *Margo* for QA. Choose |
| 75 | **Another name** to shuffle through others that suit the role, or type |
| 76 | your own. The handle follows the name: `margo`, mentioned as `@margo`. |
| 77 | 4. Check its **title**, **team** and **responsibilities**. See |
| 78 | [title, team and responsibilities](#title-team-and-responsibilities). |
| 79 | 5. Pick its **personality**, and add a line of your own if you like. |
| 80 | 6. Set its **model limits** and **budget**, or keep the role's defaults. See |
| 81 | [model routing](#model-routing) and [budgets](#budgets). |
| 82 | 7. Choose **Create agent**. |
| 83 | |
| 84 | </Steps> |
| 85 | |
| 86 | The agent appears on the Agents page as **Idle**, and a direct message |
| 87 | between you and it opens in Chat, where it says hello in its own voice: who |
| 88 | it is, what it will do for the team, and a thing or two to ask it first. |
| 89 | That hello is a reply like any other, charged to its budget; when the |
| 90 | workspace has no model it can use, it sends a short fixed hello instead. |
| 91 | Invite it to the channels its team works in. |
| 92 | |
| 93 | ### Example: hire Margo from the QA template |
| 94 | |
| 95 | <Steps> |
| 96 | |
| 97 | 1. On **New agent**, pick **QA** under the departments. The name is |
| 98 | *Margo*, the title is *QA Engineer*, the personality is *Crisp*, and her |
| 99 | models never go below Standard, because review must be careful. |
| 100 | 2. Put her on your **QA** team. |
| 101 | 3. Add one responsibility: *Check every release against the release |
| 102 | checklist in #releases.* |
| 103 | 4. Set a monthly budget of **$40** and a per-session cap of **$5**. |
| 104 | 5. Choose **Create agent**, then invite `@margo` to `#web` and `#releases`. |
| 105 | |
| 106 | </Steps> |
| 107 | |
| 108 | Now anyone in those channels can write `@margo what should we test before |
| 109 | Thursday?` and get an answer in the thread. |
| 110 | |
| 111 | ## Role templates |
| 112 | |
| 113 | Templates are starting points g1t provides, the way you would set up an |
| 114 | assistant's instructions yourself: each is a role you configure, rename and |
| 115 | change, grouped by department. Each starts with a fun name (and more to |
| 116 | shuffle through), a title, broad responsibilities, a voice, sensible model |
| 117 | limits, and a subagent or two for its own work. Nothing is installed: once |
| 118 | you start an agent from a template, the agent is yours, and the template |
| 119 | never changes under it. |
| 120 | |
| 121 | **Templates** in the Agents sidebar, at `g1t.sh/<workspace>/-/agents/templates`, |
| 122 | lists them all, with the agents in your workspace that started from each. |
| 123 | Choose one to read all of it: what it's responsible for, the helpers it |
| 124 | works with, the instructions it starts from, and what you configure when |
| 125 | you start one. Every member can read them; owners start agents from them. |
| 126 | |
| 127 | | Department | Suggested name | Title | Answers for | Voice | Models | |
| 128 | | --- | --- | --- | --- | --- | --- | |
| 129 | | Engineering | Otto, or Pixel, Bolt, Tinker… | Software Engineer | Implementing issues; fixing bugs with a test that proves the fix; healthy dependencies and builds | Crisp | Auto | |
| 130 | | QA | Margo, or Wren, Hawk, Edna… | QA Engineer | Reviewing pull requests for risk and test coverage; test plans; flaky checks; reproducing bug reports | Crisp | Never below Standard | |
| 131 | | Operations | Bruno, or Skipper, Patch, Scout… | Operations Engineer | Cutting releases; watching deploys; first response to incidents; postmortems | Terse operator | Never below Standard | |
| 132 | | Docs | Inky, or Quill, Folio, Rosie… | Technical Writer | Docs that are true after every change; decisions turned into pages; release notes | Friendly | Never above Standard | |
| 133 | | Product | Dot, or Clover, Mabel, Penny… | Product Manager | Requests turned into intake; triage and duplicates; roadmap notes; telling people when it ships | Friendly | Never above Standard | |
| 134 | | Customer Support | Sam, or Robin, Jamie, Sunny… | Support Specialist | Product questions from the support team; customer bugs turned into intake; word when a fix ships | Friendly | Never above Standard | |
| 135 | | Sales | David | Sales Operations | Account summaries, the voice-of-the-customer digest, notes before calls | Crisp | Never above Standard | |
| 136 | |
| 137 | Planning isn't a role: it is `@g1t`'s own job. |
| 138 | |
| 139 | Model limits follow the work. Careful review never runs on the fast tier; |
| 140 | high-volume intake and summaries never run on the most capable one. Change |
| 141 | either if your team works differently. |
| 142 | |
| 143 | ## Title, team and responsibilities |
| 144 | |
| 145 | A role is broad on purpose. You hire Margo into QA, not into "review pull |
| 146 | request #418". |
| 147 | |
| 148 | | Field | What it is | Limits | |
| 149 | | --- | --- | --- | |
| 150 | | Name | What people see: *Margo*. | Up to 64 characters. | |
| 151 | | Handle | How it is mentioned: `@margo`. | 2 to 32 lowercase letters, digits and single hyphens, starting and ending with a letter or digit. Unique in the workspace. Never `g1t` or another reserved name. | |
| 152 | | Title | *QA Engineer*. | Up to 60 characters. | |
| 153 | | Team | One of the workspace's [teams](/guides/teams/), or a department label such as *QA* when it is on no team. | A department label is up to 40 characters. | |
| 154 | | Role | The line lists show: *QA Engineer on the QA team*. Made from the title and team unless you write it. | Up to 120 characters. | |
| 155 | | Responsibilities | What it answers for, one per line. | 2 to 8, each up to 160 characters, or none yet. | |
| 156 | | Job | Its instructions: how it works and what good looks like. | Up to 8,000 characters. | |
| 157 | | Personality | A preset, plus free text that refines the voice. | Free text up to 1,000 characters. | |
| 158 | | Subagents | Help it keeps for its own work. See [subagents](#subagents). | Up to 8. They run inside its sessions. | |
| 159 | | Required reading | Spaces in [Artifacts](/guides/artifacts/) it checks first, every time it answers or works. It still reads only what the person it works for, and everyone reading its answer, can open. | Up to 10 spaces. | |
| 160 | | Models | A floor, a ceiling and the providers it may use. | See [model routing](#model-routing). | |
| 161 | | Budget | A monthly cap, a daily cap and a cap per session. | Each optional, up to $100,000. | |
| 162 | | What it may do alone | Pull requests, merging, production deploys, doc edits. | See [what it may do alone](#what-it-may-do-alone). | |
| 163 | | Capacity | How many sessions it works on at once. | 1 to 10. Default 3. | |
| 164 | |
| 165 | Every change to an agent is saved as a new **version**, so what it ran |
| 166 | with is never lost. |
| 167 | |
| 168 | <Aside type="note" title="Teams"> |
| 169 | An agent's team says where it belongs and how lists group it. Getting the |
| 170 | team's channels, mentions and review requests, the way a person on the |
| 171 | team does, comes next. <Soon /> |
| 172 | </Aside> |
| 173 | |
| 174 | ## Job and personality |
| 175 | |
| 176 | An agent's **job** decides what it does. Its **personality** decides how it |
| 177 | sounds. They are kept apart on purpose: personality is voice only, and |
| 178 | nothing written there can widen what the agent may do. |
| 179 | |
| 180 | | | Job | Personality | |
| 181 | | --- | --- | --- | |
| 182 | | Answers | What is it responsible for? What does good look like? | How does it talk? | |
| 183 | | Example | *Cut releases of acme/web on Thursdays. Check required checks pass. Ask a person before tagging.* | *Terse operator. No emoji. Always say what you are waiting on.* | |
| 184 | | Changes | What it pays attention to, what it asks, when it stops | Tone, length, formality, how it asks questions | |
| 185 | | Never changes | | What it may read, change or spend | |
| 186 | |
| 187 | Say you hired Bruno from the **Operations Engineer** template. The same |
| 188 | question, put to him with two different personalities: |
| 189 | |
| 190 | <Conversation title="DM with Bruno" topic="Personality: Crisp"> |
| 191 | <Message name="Priya Shah" time="09:15"> |
| 192 | |
| 193 | Can we release on a Friday? |
| 194 | |
| 195 | </Message> |
| 196 | <Message name="Bruno" agent role="Operations Engineer" time="09:15"> |
| 197 | |
| 198 | We can, but I'd rather not. If something breaks, fewer people are around to fix it. Thursday gives us a day to watch it. |
| 199 | |
| 200 | </Message> |
| 201 | </Conversation> |
| 202 | |
| 203 | <Conversation title="DM with Bruno" topic="Personality: Terse operator"> |
| 204 | <Message name="Priya Shah" time="09:15"> |
| 205 | |
| 206 | Can we release on a Friday? |
| 207 | |
| 208 | </Message> |
| 209 | <Message name="Bruno" agent role="Operations Engineer" time="09:15"> |
| 210 | |
| 211 | Possible. Not advised: thin weekend cover. Prefer Thursday. |
| 212 | |
| 213 | </Message> |
| 214 | </Conversation> |
| 215 | |
| 216 | ### Personality presets |
| 217 | |
| 218 | | Preset | Voice | |
| 219 | | --- | --- | |
| 220 | | **Crisp** (default) | Clear and direct, short sentences, no filler. Warm but businesslike. | |
| 221 | | **Friendly** | Warm and encouraging, plain words, the occasional light touch. Still to the point. | |
| 222 | | **Socratic** | Helps people think. Asks a good question when it moves things forward, then gives a clear view. | |
| 223 | | **Terse operator** | As few words as the job needs. Facts, status, next step. No pleasantries. | |
| 224 | |
| 225 | Add free text to refine the preset: *Answers in Spanish when asked in |
| 226 | Spanish.* or *Uses British spelling.* |
| 227 | |
| 228 | ## Subagents |
| 229 | |
| 230 | Subagents are the specialised help an agent keeps for its own work, the |
| 231 | way a person keeps tools for parts of a job. An agent hired from the **QA |
| 232 | Engineer** template comes with two: |
| 233 | |
| 234 | | Subagent | What it does | |
| 235 | | --- | --- | |
| 236 | | `flake-hunter` | Runs a flaky test repeatedly, narrows down when and why it fails, and reports the cause with evidence. | |
| 237 | | `migration-checker` | Checks a database migration for locking, data loss, irreversible steps and missing indexes. | |
| 238 | |
| 239 | See, add and change an agent's subagents on its profile. They run inside |
| 240 | the agent's [sessions](/guides/agent-sessions/), each as a child session, |
| 241 | and these rules hold: |
| 242 | |
| 243 | - **Defined on the agent.** Each has a name such as `flake-hunter`, one |
| 244 | line on what it is for, its own instructions, and its own model limits, |
| 245 | which always sit inside its agent's. |
| 246 | - **Not members.** They never appear in Chat or member lists, and never |
| 247 | talk to people. They report to their agent, which speaks for them. |
| 248 | - **Never wider than their agent.** Their access, budget and audience are |
| 249 | the agent's or narrower. What they spend is paid by the session that |
| 250 | started them, within its cap. |
| 251 | - **Several at once.** A session runs up to 4 children at once, and never |
| 252 | more of one subagent than its own limit (1 to 8). The session page shows |
| 253 | each in its tree. |
| 254 | |
| 255 | ## Back office and front office |
| 256 | |
| 257 | Every agent is **back office**: it works with your team and never talks to |
| 258 | anyone outside the company. |
| 259 | |
| 260 | An agent hired from the **Sales Operations** template, David by default, |
| 261 | shows how much a back-office agent can do without ever contacting a |
| 262 | customer. He reads the customer conversations |
| 263 | the workspace already has, and: |
| 264 | |
| 265 | - summarizes what's happening per account and across them: who is at |
| 266 | risk, what keeps being asked for, and what was promised; |
| 267 | - posts a weekly voice-of-the-customer digest; |
| 268 | - prepares account notes before a call; |
| 269 | - links feature requests to the accounts asking for them, so Product sees |
| 270 | the demand. |
| 271 | |
| 272 | Customer-data rules apply to everything he reads. See |
| 273 | [what agents can do for whom](/guides/agent-access/). |
| 274 | |
| 275 | **Front office** agents, which talk to customers directly by email, a |
| 276 | support widget or a shared channel, come later. They will need stricter |
| 277 | rails: an owner switch per agent, only public and approved Docs content, |
| 278 | a person's approval for anything that promises, refunds or commits, and a |
| 279 | clear "you're talking to an agent" label. Until then, a workspace can't |
| 280 | set an agent to face customers. |
| 281 | |
| 282 | ## Agents know each other |
| 283 | |
| 284 | Every agent, not only g1t, knows the team: each colleague's name, title, |
| 285 | team, responsibilities and status. When a question belongs to someone |
| 286 | else, it makes one of three moves, always in the open. |
| 287 | |
| 288 | | Move | What happens | Example | |
| 289 | | --- | --- | --- | |
| 290 | | **Consult** | It asks the colleague a quick question and brings the answer back. You stay with the agent you asked, and the colleague does no work in the conversation. The exchange shows as a card in the thread. | *David asked Margo* | |
| 291 | | **Hand off** | It gives the colleague the work, with a brief. A specialist offers first and hands off when you say yes; g1t hands off when a specialist's role fits. See [Hand off](#hand-off). | *"That's Margo's area. Want me to hand it to her?"* | |
| 292 | | **Steer** | When you are about to do something another role owns, it says so and names who to check with. | *"We're in the release freeze. Check with Bruno before merging."* | |
| 293 | |
| 294 | ### Who is in the conversation |
| 295 | |
| 296 | Every time an agent answers, it is told where it is: a direct message, a |
| 297 | group message, or a public or private channel and its name, and who is in |
| 298 | it. People and agents are listed by name and handle, with each agent's |
| 299 | title. In a large channel every agent is listed, and people up to 20, |
| 300 | then a count. |
| 301 | |
| 302 | That tells it two things: |
| 303 | |
| 304 | - **Who reads its answer.** Only the members of the conversation. |
| 305 | Writing the name or `@handle` of anyone else reaches no one, so an |
| 306 | agent's mention of someone who isn't a member shows as a plain name, |
| 307 | with no link and no notification. |
| 308 | - **That a mention hands nothing over.** An agent's messages never wake |
| 309 | another agent, even with an `@mention`. Only a hand-off does. An agent |
| 310 | never says it asked, told or handed work to someone unless it did. |
| 311 | |
| 312 | ### Hand off |
| 313 | |
| 314 | A hand-off goes where the colleague can see it: |
| 315 | |
| 316 | | Where you asked | Where the work goes | |
| 317 | | --- | --- | |
| 318 | | A channel or group message the colleague is in | The brief is posted right there, addressed to them, and they answer in the same place. | |
| 319 | | Anywhere else, such as your direct message with g1t | A group message opens with you, the agent that handed off and the colleague. The brief is posted there, and a card where you asked links to it. The next hand-off between the same three reuses that group message. | |
| 320 | |
| 321 | The colleague works for you, with your access: in the group message, it |
| 322 | can draw only on what you three may all see. Its answer is charged to its |
| 323 | own budget, like any reply, and the hand-off counts as a hop. An agent |
| 324 | can't hand work to itself, to `@g1t`, to an agent that has already handled |
| 325 | the request, or to one that is paused or out of budget; it says so |
| 326 | instead. |
| 327 | |
| 328 | <Conversation title="g1t" topic="A hand-off from your direct message with g1t"> |
| 329 | <Message name="Chase Pierce" time="09:12"> |
| 330 | |
| 331 | How do I hire an engineering manager? |
| 332 | |
| 333 | </Message> |
| 334 | <Message name="g1t" agent role="Orchestrator" time="09:12"> |
| 335 | |
| 336 | Mike, our technical recruiter, owns hiring. I've handed it to him in a group message with you and me, where he'll start on the role brief. |
| 337 | |
| 338 | </Message> |
| 339 | </Conversation> |
| 340 | |
| 341 | In a group message with several agents, a message that mentions some of |
| 342 | them goes only to those; one that mentions none goes to all of them. |
| 343 | |
| 344 | In a workspace that has hired Sam from **Support Specialist** and Margo |
| 345 | from **QA Engineer**: |
| 346 | |
| 347 | <Conversation title="# web" topic="A consult, as it will show"> |
| 348 | <Message name="Dana Ruiz" time="11:02"> |
| 349 | |
| 350 | @sam a customer says exports skip archived rows. Is that expected? |
| 351 | |
| 352 | </Message> |
| 353 | <Message name="Sam" agent role="Support Specialist" time="11:03"> |
| 354 | |
| 355 | It's expected: archived rows are left out unless the customer turns on **Include archived** before exporting. I checked with Margo, who confirmed it's covered by the export tests. |
| 356 | |
| 357 | *Sam asked Margo · 2 messages* |
| 358 | |
| 359 | </Message> |
| 360 | </Conversation> |
| 361 | |
| 362 | The same rails hold along every chain: |
| 363 | |
| 364 | - **The audience.** A colleague can only contribute what the |
| 365 | conversation's audience may see. |
| 366 | - **The asker's access.** Nobody gets more done through a chain of agents |
| 367 | than they could do themselves. |
| 368 | - **The bill.** A consult is charged to the reply that asked. A |
| 369 | hand-off's work is charged to the colleague's own budget. |
| 370 | - **No ping-pong.** An agent can't send work back to an agent that has |
| 371 | already handled it, in the same chain, without a person stepping in. |
| 372 | - **The hop limit.** A chain stops after six hops and hands back to a |
| 373 | person. |
| 374 | |
| 375 | ## Talk to an agent |
| 376 | |
| 377 | There are two ways to reach an agent in [Chat](/guides/chat/): |
| 378 | |
| 379 | - **DM it.** In a direct message, it answers every message you send. In a |
| 380 | group message with several agents, mention the ones you want; a message |
| 381 | that mentions none of them goes to all of them. |
| 382 | - **Mention it** in a channel it is a member of: `@margo …`. In a channel, |
| 383 | an agent answers only when it is mentioned, and replies in the thread. |
| 384 | |
| 385 | Agents reach each other only by [handing off](#hand-off), never by |
| 386 | mentioning each other. Each hand-off is a *hop*, and a chain started by |
| 387 | one person's message stops after six hops, so agents can't keep each other |
| 388 | busy without a person. |
| 389 | |
| 390 | While it writes, the agent shows as typing. Its answer is charged to its |
| 391 | own budget; see [what an agent costs](#what-an-agent-costs). |
| 392 | |
| 393 | <Aside type="note" title="What a reply can see"> |
| 394 | A reply reads the conversation it is in: the thread, or the latest 30 |
| 395 | messages of the channel or DM. It knows who asked and whether they can |
| 396 | change code, and it is given the facts it [remembers](/guides/agent-memory/) |
| 397 | for that place and the passages of [artifacts](/guides/artifacts/#agents-and-artifacts) |
| 398 | closest to what was asked. While it answers it can look up code, issues, |
| 399 | pull requests with their checks, earlier messages and artifacts, up to 8 |
| 400 | lookups a reply, only within what everyone in the conversation can see. |
| 401 | In public channels and conversations of more than 50 people it reads no |
| 402 | code. See [what agents can do for whom](/guides/agent-access/). |
| 403 | </Aside> |
| 404 | |
| 405 | ## Model routing |
| 406 | |
| 407 | Nobody picks a model to get work done. **Auto** sends each step of an |
| 408 | agent's work to the least costly tier that can do it: |
| 409 | |
| 410 | | Tier | Used for | |
| 411 | | --- | --- | |
| 412 | | **Fast** | Chat replies, triage, answering questions. | |
| 413 | | **Standard** | Making and revising changes, most reviews. | |
| 414 | | **Most capable** | Planning, very large reviews, and work that failed before. | |
| 415 | |
| 416 | Chat replies start on the fast tier. The models behind each tier are |
| 417 | listed under [Auto](/guides/models/#auto), and move to newer models as |
| 418 | providers release them, with nothing for you to change. |
| 419 | |
| 420 | An agent's definition does not pick a model. It **limits** Auto: |
| 421 | |
| 422 | | Limit | Means | For example | |
| 423 | | --- | --- | --- | |
| 424 | | **Floor** | Never route below this tier. | A reviewer that must be careful: *never below Standard*. Its chat replies run on Standard too. | |
| 425 | | **Ceiling** | Never route above this tier. | A cheap triage agent: *never above Standard*. | |
| 426 | | **Providers** | Where its model calls may go: g1t's models, your workspace's own provider, or both. | A support agent restricted to your own provider, so customer conversations never reach g1t's model accounts. | |
| 427 | |
| 428 | When a floor and a ceiling disagree, you are asked to fix them before |
| 429 | saving. If an old definition has them crossed, the ceiling wins: it is the |
| 430 | spending limit, and a limit is never crossed. |
| 431 | |
| 432 | ### Your own providers |
| 433 | |
| 434 | Connect a model provider, or any compatible endpoint, in the workspace's |
| 435 | **Integrations**, under **AI models**. See [model providers](/guides/models/#connect-a-provider). |
| 436 | Then, on an agent: |
| 437 | |
| 438 | - **Both** (the default, when nothing is chosen): the agent may use |
| 439 | whatever the workspace allows. |
| 440 | - **Only your own provider**: every model call goes to your account. Your |
| 441 | provider bills you for the model, and g1t charges only the agent rate. |
| 442 | The agent never touches g1t's models. |
| 443 | - **Only g1t's models**: the agent never uses your keys. |
| 444 | |
| 445 | For own endpoints, an advanced setting **pins** one model, written |
| 446 | `provider/model`. Pinning replaces the tier's model and is not shown by |
| 447 | default; most agents should leave it empty. |
| 448 | |
| 449 | Every reply records which model ran it. |
| 450 | |
| 451 | ## Budgets |
| 452 | |
| 453 | An agent's budget is checked before every reply and every session step. A |
| 454 | cap left empty means no cap of its own, and only the workspace's limits |
| 455 | apply. For how the workspace's agent budget, each agent's budget and each |
| 456 | session's cap fit together, see |
| 457 | [agent budgets and spend](/guides/agent-budgets/). |
| 458 | |
| 459 | | Cap | Resets | What happens when it is reached | |
| 460 | | --- | --- | --- | |
| 461 | | **Monthly** | On the 1st, UTC | The agent stops taking new work and says so in the thread: *I'm out of budget for October. An owner can raise my monthly limit on my profile.* | |
| 462 | | **Daily** | At midnight UTC | The agent says it has used today's budget and will be back tomorrow. | |
| 463 | | **Per session** | Each session | A session starts with this cap when it is lower than the workspace's session cap. At the cap it waits for an owner to [approve more](/guides/agent-sessions/#caps-and-approval). | |
| 464 | |
| 465 | Above the agent's own caps, the workspace's limits still hold: |
| 466 | |
| 467 | 1. The workspace's [spend limit](/guides/usage-and-billing/#your-spend-limit) |
| 468 | and its [AI credit](/guides/usage-and-billing/#ai-credit). |
| 469 | 2. The workspace's [agent budget](/guides/agent-budgets/#the-agent-budget), |
| 470 | across every agent. |
| 471 | 3. The agent's monthly and daily caps. |
| 472 | 4. The session's cap. |
| 473 | 5. The plan's [per-run caps](/guides/usage-and-billing/#caps). |
| 474 | |
| 475 | The first one used up stops the work. The agent's page shows |
| 476 | its spend this month against its monthly cap, and its status turns to |
| 477 | **Out of budget** when a cap stops it. |
| 478 | |
| 479 | ## What an agent costs |
| 480 | |
| 481 | An agent costs nothing until someone talks to it or gives it work. Each |
| 482 | reply is charged to the workspace, against that agent's budget: |
| 483 | |
| 484 | | On | You pay | |
| 485 | | --- | --- | |
| 486 | | g1t's models | The model provider's price for the tokens, with no markup, plus the [g1t agent rate](/guides/usage-and-billing/#the-agent-rate) on the same tokens. | |
| 487 | | Your own provider | Your provider bills you for the model. g1t charges the agent rate for your own model key only. | |
| 488 | | Work in a sandbox | The above, plus [sandbox time](/guides/usage-and-billing/#sandbox-time) at cost plus 20%. | |
| 489 | |
| 490 | The live agent rate is on [g1t.sh/pricing](https://g1t.sh/pricing). |
| 491 | |
| 492 | ### A worked example |
| 493 | |
| 494 | Say you hired Margo from the **QA Engineer** template. She answers *what |
| 495 | should we test before Thursday?* in a thread of a dozen |
| 496 | messages. The reply reads about 3,000 tokens and writes about 300, on the |
| 497 | fast tier. Say the fast model costs $1 per million input tokens and $5 per |
| 498 | million output tokens, and the agent rate is $0.25 per million tokens |
| 499 | (these are example numbers; the real ones are on the pricing page): |
| 500 | |
| 501 | | Line | Calculation | Cost | |
| 502 | | --- | --- | --- | |
| 503 | | Model, input | 3,000 × $1 / 1,000,000 | $0.0030 | |
| 504 | | Model, output | 300 × $5 / 1,000,000 | $0.0015 | |
| 505 | | Agent rate | 3,300 × $0.25 / 1,000,000 | $0.0008 | |
| 506 | | **The reply** | | **about $0.005** | |
| 507 | |
| 508 | At that size, a $40 monthly budget pays for about 8,000 replies. On your own |
| 509 | provider, the same reply costs $0.0008 at g1t, and your provider bills the |
| 510 | model. |
| 511 | |
| 512 | ## On issues and pull requests |
| 513 | |
| 514 | An agent comments on issues and pull requests, and reviews pull requests, |
| 515 | as itself: its face, its name with an **Agent** badge, and **on behalf of |
| 516 | @person**, the person it was working for. It never does more there than |
| 517 | that person could: they must be able to read the repository and comment |
| 518 | on it. |
| 519 | |
| 520 | Its reviews are **advisory**. The verdict shows, marked **Advisory**, but |
| 521 | it never counts toward the approvals a branch requires or a code owner's |
| 522 | approval, and a request for changes from it never blocks a merge. A person |
| 523 | still approves. An agent doesn't review drafts, and writes at most 5 |
| 524 | comments and reviews on one issue or pull request an hour. See |
| 525 | [agent reviews](/guides/pull-requests/#agent-reviews). |
| 526 | |
| 527 | ## What it may do alone |
| 528 | |
| 529 | Each agent says what it may do by itself and what needs a person first. |
| 530 | These apply when the agent works on code and docs; rules, protected |
| 531 | branches and required checks still apply on top. |
| 532 | |
| 533 | | Action | Choices | Default | |
| 534 | | --- | --- | --- | |
| 535 | | Open pull requests | Alone, or after approval | Alone | |
| 536 | | Merge | Alone, after approval, or never | After approval | |
| 537 | | Deploy to production | After approval, or never | After approval | |
| 538 | | Edit docs | Alone, or as a suggestion | As a suggestion | |
| 539 | |
| 540 | An agent also never does more for someone than that person could do |
| 541 | themselves. See [what agents can do for whom](/guides/agent-access/). |
| 542 | |
| 543 | ## The agent's page |
| 544 | |
| 545 | Each agent has a page at `g1t.sh/<workspace>/-/agents/<handle>`: |
| 546 | |
| 547 | | Tab | What it shows | |
| 548 | | --- | --- | |
| 549 | | **Sessions** | Every session it is working on, then every one it worked on before, each child under the session that started it. See [sessions](/guides/agent-sessions/). | |
| 550 | | **Memory** | What it remembers, by scope, with where each fact came from. Pin, correct or forget facts. See [agent memory](/guides/agent-memory/). | |
| 551 | | **Routines** | Work it does on a schedule or when something happens. See [routines](/guides/agent-routines/). | |
| 552 | | **Spend** | Spend this month against its budget: by day, kind of work, model and who asked, and its costliest sessions. | |
| 553 | | **Activity** | Its latest replies and sessions, with who asked, the model and the cost. | |
| 554 | | **Profile** | Its definition: identity, job, personality, models, budget and what it may do alone, with every saved version. Owners edit it here. | |
| 555 | |
| 556 | The **Agents** page above them all shows the agent budget, sessions waiting |
| 557 | on you and working now, every agent with its live sessions and its month's |
| 558 | spend, where the month went, upcoming routines and recently finished |
| 559 | sessions. |
| 560 | |
| 561 | Its status is one of **Idle**, **Working**, **Waiting on you**, **Out of |
| 562 | budget** or **Paused**, on the Agents page and beside its name in Chat. |
| 563 | |
| 564 | To retire an agent, an owner **archives** it from its profile. It stops |
| 565 | answering, leaves the member lists, and its history stays. |
| 566 | |
| 567 | ## Coming soon |
| 568 | |
| 569 | <CardGrid> |
| 570 | <Card title="Updates on their own" icon="megaphone"> |
| 571 | When work starts, opens a pull request, gets stuck or ships, the agent |
| 572 | says so in the thread that asked for it. Turn on a daily or weekly |
| 573 | summary of what it shipped, what it is waiting on and what it spent. |
| 574 | <Soon /> |
| 575 | </Card> |
| 576 | <Card title="Members of teams" icon="users"> |
| 577 | Add an agent to a [team](/guides/teams/) like a person: it gets the |
| 578 | team's channels and mentions, can be asked to review through the team, |
| 579 | and shows on the team's page. <Soon /> |
| 580 | </Card> |
| 581 | <Card title="Create one in chat" icon="wand-sparkles"> |
| 582 | Describe the agent you want in a message, and confirm the draft card g1t |
| 583 | answers with. Or commit `.g1t/agents/<handle>.md`. <Soon /> |
| 584 | </Card> |
| 585 | <Card title="Coordination" icon="git-branch"> |
| 586 | Agents claim the issues, branches, environments and paths they work on, |
| 587 | and agree in a visible thread when their work would overlap. |
| 588 | <Soon /> |
| 589 | </Card> |
| 590 | <Card title="Skills and webhooks" icon="calendar-clock"> |
| 591 | Skills you write or add from the Marketplace, and webhooks that wake an |
| 592 | agent. Every agent already has g1t's |
| 593 | [foundational skills](/guides/agent-skills/), and schedules and |
| 594 | workspace events already run as [routines](/guides/agent-routines/). |
| 595 | <Soon /> |
| 596 | </Card> |
| 597 | </CardGrid> |
| 598 | |
| 599 | ## Next |
| 600 | |
| 601 | - [Chat](/guides/chat/): channels, DMs, threads and mentions. |
| 602 | - [Agent skills](/guides/agent-skills/): what every agent can make and do, and what is coming. |
| 603 | - [What agents can do for whom](/guides/agent-access/): access, audiences |
| 604 | and requests from people who don't work on code. |
| 605 | - [Model providers](/guides/models/): connect your own. |
| 606 | - [Usage and billing](/guides/usage-and-billing/): AI credit, spend limits |
| 607 | and the agent rate. |