Skip to content

Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action decided before it happens and recorded after. Bring any AG-UI agent.

Run every Bot on a current model (#168)

David McKayf74d116878b3
diff --git a/.env.example b/.env.example
index cf05680a..f8a63e24 100644
--- a/.env.example
+++ b/.env.example
@@ -125,7 +125,7 @@ OPENAI_API_KEY=
#
# OPENAI_BASE_URL=https://gateway.internal/v1
# OPENAI_API_KEY=...
-# BOT_MODEL=openai/gpt-4o
+# BOT_MODEL=openai/gpt-5.6-terra
#
# OPENAI_BASE_URL=
@@ -141,11 +141,12 @@ OPENAI_API_KEY=
# ANTHROPIC_API_KEY=
# GOOGLE_API_KEY=
-# Which model. Defaults per provider: gpt-5.5, claude-sonnet-4-5, gemini-2.5-flash.
-# BOT_MODEL=gpt-5.5
+# Which model. Defaults per provider: gpt-5.6-terra, claude-sonnet-4-5, gemini-2.5-flash.
+# OpenAI's 5.6 tiers are sol (most capable), terra (the default here) and luna (cheapest).
+# BOT_MODEL=gpt-5.6-terra
-# OpenAI only. Its newer models require the Responses API, which the framework Bot handles and the
-# proof-of-concept one cannot.
+# OpenAI only, and rarely needed: the framework Bot turns the Responses API on by itself for models
+# that require it. Set it when you are using a model this build has not heard of that needs it too.
# BOT_RESPONSES_API=false
# The Bot computer. Absent means the feature is off and its routes are not mounted.
@@ -237,8 +238,10 @@ MANAGED_AGENT_TOKEN=
# proof of concept, and is reached the same way: point MANAGED_AGENT_AG_UI_URL at it, or add it as a
# Bot of its own in the tenant package or at /agents.
-# Which model the Bots use. agent-bot speaks /v1/chat/completions and cannot use gpt-5.6-*, which
-# require the Responses API. agent-langgraph can: set BOT_RESPONSES_API=true and give it a 5.6 model.
+# Which model the Bots use. agent-langgraph runs gpt-5.6-terra and switches to the Responses API by
+# itself, because 5.6 rejects function tools on /v1/chat/completions. agent-bot speaks that endpoint
+# by hand and stays on gpt-5.5: the alternative there is reasoning_effort 'none', and a Bot that has
+# to decide when to ask a person for help should not be the one with its reasoning turned off.
# BOT_RESPONSES_API=false
# One computer per Bot. Unset, every Bot shares the computer at AGENT_COMPUTER_URL, suitable on a
diff --git a/agent-langgraph/src/index.ts b/agent-langgraph/src/index.ts
index 3d183e1a..b8a697d6 100644
--- a/agent-langgraph/src/index.ts
+++ b/agent-langgraph/src/index.ts
@@ -24,9 +24,10 @@ import { toLangChainMessages } from "./history";
* server changes.
*
* The model API stops being ours. `agent-bot` speaks `/v1/chat/completions` by hand, which is why
- * gpt-5.6 cannot be used there without rewriting its streaming loop: those models reject function
- * tools on that endpoint and require the Responses API. Here it is `useResponsesApi`, one line,
- * because the migration belongs to the people who maintain the integration.
+ * gpt-5.6 costs it a rewritten streaming loop: those models reject function tools on that endpoint
+ * unless reasoning is turned off, and turning reasoning off on a Bot that has to decide when to ask
+ * a person for help is the wrong trade. Here it is `useResponsesApi`, one line, because the
+ * migration belongs to the people who maintain the integration.
*
* The tool loop still runs on the client, exactly as it does in `agent-bot`: a Bot's actions happen
* on a browser the person is watching, and the surface remains the place that executes those tools.
@@ -66,8 +67,17 @@ const PROVIDER = (process.env.BOT_PROVIDER ?? "openai").toLowerCase();
* missing configuration.
*/
const MODEL = process.env.BOT_MODEL?.trim() || defaultModelFor(PROVIDER);
-/** OpenAI only. Its newer models require the Responses API, which the integration handles. */
-const USE_RESPONSES_API = process.env.BOT_RESPONSES_API === "true";
+/**
+ * OpenAI only. Its newer models require the Responses API, which the integration handles.
+ *
+ * Inferred from the model rather than left to a separate switch. `gpt-5.6-*` rejects function tools
+ * on `/v1/chat/completions`, so a deployment that set `BOT_MODEL` to one and did not also know about
+ * this flag got a Bot that started, looked healthy, and failed on its first tool call. The switch is
+ * still honoured, so a model this list has not heard of can be told to use it.
+ */
+const NEEDS_RESPONSES_API = /^gpt-5\.[6-9]|^gpt-[6-9]/.test(MODEL);
+const USE_RESPONSES_API =
+ process.env.BOT_RESPONSES_API === "true" || NEEDS_RESPONSES_API;
/**
* OpenAI only, and the same variable the API server reads for its built-in agents.
*
@@ -90,7 +100,7 @@ const GOOGLE_BASE_URL =
function defaultModelFor(provider: string): string {
if (provider === "anthropic") return "claude-sonnet-4-5";
if (provider === "google") return "gemini-2.5-flash";
- return "gpt-5.5";
+ return "gpt-5.6-terra";
}
/**
diff --git a/docker-compose.yml b/docker-compose.yml
index 4c67db89..bbe2a643 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -177,7 +177,7 @@ services:
# Unset means OpenAI. Set, it is any endpoint speaking the same API, and BOT_MODEL is sent
# to it verbatim.
OPENAI_BASE_URL: ${OPENAI_BASE_URL:-}
- BOT_MODEL: ${BOT_MODEL:-gpt-5.5}
+ BOT_MODEL: ${BOT_MODEL:-gpt-5.6-terra}
healthcheck:
test: ["CMD-SHELL", "bun -e \"await fetch('http://localhost:4200/health')\""]
interval: 10s
@@ -204,7 +204,7 @@ services:
ANTHROPIC_BASE_URL: ${ANTHROPIC_BASE_URL:-}
GOOGLE_API_KEY: ${GOOGLE_API_KEY:-}
GOOGLE_GENERATIVE_AI_BASE_URL: ${GOOGLE_GENERATIVE_AI_BASE_URL:-}
- BOT_MODEL: ${BOT_MODEL:-gpt-5.5}
+ BOT_MODEL: ${BOT_MODEL:-gpt-5.6-terra}
BOT_RESPONSES_API: ${BOT_RESPONSES_API:-false}
# Where this Bot runs a tool: back through the deployment that granted it, never at the vendor.
# `host.docker.internal` because the API server runs on the host, not in this network.
diff --git a/docs/configuration.md b/docs/configuration.md
index 95adf337..4943c734 100644
--- a/docs/configuration.md
+++ b/docs/configuration.md
@@ -93,7 +93,7 @@ A gateway that fronts several providers behind one key is addressed the usual wa
```sh
OPENAI_BASE_URL=https://gateway.internal/v1
OPENAI_API_KEY=...
-BOT_MODEL=openai/gpt-4o
+BOT_MODEL=openai/gpt-5.6-terra
```
and in the tenant package, where the name is namespaced the same way:
@@ -102,7 +102,7 @@ and in the tenant package, where the name is namespaced the same way:
model:
provider: openai
credential_secret_ref: openai-api-key
- default_model: openai/gpt-4o
+ default_model: openai/gpt-5.6-terra
```
Most gateways publish a model list, which is the way to check a name before configuring it.
@@ -366,7 +366,7 @@ column nor `users.groups` is the wrong shape for it.
model:
provider: openai
credential_secret_ref: openai-api-key
- default_model: gpt-4.1
+ default_model: gpt-5.6-terra
```
`provider` must be `openai`. `credential_secret_ref` is a reference to a stored credential, not a credential value. `default_model` is passed through as written, so an OpenAI-compatible endpoint reached through `OPENAI_BASE_URL` takes the name that endpoint publishes.
diff --git a/examples/fintech/model.yaml b/examples/fintech/model.yaml
index 5bdce6a2..2c0d518a 100644
--- a/examples/fintech/model.yaml
+++ b/examples/fintech/model.yaml
@@ -1,4 +1,4 @@
model:
provider: openai
credential_secret_ref: openai-api-key
- default_model: gpt-4.1
+ default_model: gpt-5.6-terra
diff --git a/examples/langgraph-bot/src/index.ts b/examples/langgraph-bot/src/index.ts
index 6f3a1b5b..5a96f544 100644
--- a/examples/langgraph-bot/src/index.ts
+++ b/examples/langgraph-bot/src/index.ts
@@ -27,7 +27,15 @@ import { serve } from "bun";
*/
const PORT = Number.parseInt(process.env.PORT ?? "4300", 10);
-const MODEL = process.env.BOT_MODEL ?? "gpt-5.5";
+const MODEL = process.env.BOT_MODEL ?? "gpt-5.6-terra";
+/**
+ * `gpt-5.6-*` rejects function tools on `/v1/chat/completions` and needs the Responses API, which
+ * this integration speaks. Inferred from the model so setting `BOT_MODEL` alone cannot produce a
+ * Bot that starts, looks healthy, and fails on its first tool call.
+ */
+const USE_RESPONSES_API =
+ process.env.BOT_RESPONSES_API === "true" ||
+ /^gpt-5\.[6-9]|^gpt-[6-9]/.test(MODEL);
/**
* The graph's state.
@@ -131,7 +139,11 @@ function buildGraph(input: RunAgentInput) {
// the graph returns once it is finished, and the surface receives a single TEXT_MESSAGE_CONTENT
// carrying all of it, so a six-hundred-word reply is a blank conversation for eight seconds and
// then a wall of text. The tokens exist the whole time; nothing was passing them on.
- const model = new ChatOpenAI({ model: MODEL, streaming: true });
+ const model = new ChatOpenAI({
+ model: MODEL,
+ streaming: true,
+ ...(USE_RESPONSES_API ? { useResponsesApi: true } : {}),
+ });
const tools = toLangChainTools(input);
const bound = tools.length > 0 ? model.bindTools(tools) : model;
diff --git a/examples/mastra-bot/src/index.ts b/examples/mastra-bot/src/index.ts
index c5d5f715..0d99f3e8 100644
--- a/examples/mastra-bot/src/index.ts
+++ b/examples/mastra-bot/src/index.ts
@@ -17,7 +17,15 @@ import { serve } from "bun";
*/
const PORT = Number.parseInt(process.env.PORT ?? "4400", 10);
-const MODEL = process.env.BOT_MODEL ?? "gpt-5.5";
+const MODEL = process.env.BOT_MODEL ?? "gpt-5.6-terra";
+/**
+ * `gpt-5.6-*` rejects function tools on `/v1/chat/completions` and needs the Responses API, which
+ * this provider exposes as `openai.responses`. Inferred from the model so setting `BOT_MODEL` alone
+ * cannot produce a Bot that starts, looks healthy, and fails on its first tool call.
+ */
+const NEEDS_RESPONSES_API =
+ process.env.BOT_RESPONSES_API === "true" ||
+ /^gpt-5\.[6-9]|^gpt-[6-9]/.test(MODEL);
const bot = new Agent({
name: "OpenBot Mastra coworker",
@@ -28,7 +36,7 @@ const bot = new Agent({
"NEVER state what a page contains unless you have just read it with a tool in this conversation. " +
"You cannot know a page's contents from memory. If you have not read it, call the tool first, " +
"and report exactly what the tool returned.",
- model: openai(MODEL),
+ model: NEEDS_RESPONSES_API ? openai.responses(MODEL) : openai(MODEL),
});
/**
diff --git a/server/tests/copilot.test.ts b/server/tests/copilot.test.ts
index be3a4847..41d6d4b2 100644
--- a/server/tests/copilot.test.ts
+++ b/server/tests/copilot.test.ts
@@ -95,11 +95,11 @@ describe("registered Copilot agents", () => {
type: "built_in",
systemPrompt: "Be helpful.",
},
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
"openai-secret",
),
).toEqual({
- model: "openai/gpt-4.1",
+ model: "openai/gpt-5.6-terra",
// The provenance rule is unconditional, so even a Bot with no tools and no computer carries
// it. That Bot needs it most: nothing it says was read anywhere.
prompt: `Be helpful.\n\n${PROVENANCE_GUIDANCE}`,
@@ -117,7 +117,7 @@ describe("registered Copilot agents", () => {
systemPrompt: "Be helpful.",
},
],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
null,
);
const agent = agents["general-assistant"];
@@ -159,7 +159,7 @@ describe("registered Copilot agents", () => {
endpoint: "http://risk.internal/ag-ui",
},
],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
"openai-secret",
);
@@ -200,7 +200,7 @@ describe("registered Copilot agents", () => {
endpoint: "http://risk.internal/ag-ui",
},
],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
"openai-secret",
stallGuard,
);
@@ -227,7 +227,10 @@ describe("registered Copilot agents", () => {
endpoint: "http://risk.internal/ag-ui",
},
];
- const model = { provider: "openai" as const, defaultModel: "gpt-4.1" };
+ const model = {
+ provider: "openai" as const,
+ defaultModel: "gpt-5.6-terra",
+ };
const guarded = (
await buildAgents(registered, model, null, {
@@ -261,12 +264,12 @@ describe("registered Copilot agents", () => {
const first = await resolveRuntimeAgents(
async () => registered,
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
resolveModelApiKey,
);
const second = await resolveRuntimeAgents(
async () => registered,
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
resolveModelApiKey,
);
@@ -293,7 +296,7 @@ describe("registered Copilot agents", () => {
endpoint: "http://risk.internal/ag-ui",
},
],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
async () => {
resolverInvoked = true;
throw new Error("corrupt model credential");
@@ -339,7 +342,7 @@ describe("standing agent roles", () => {
await using endpoint = fakeAgUiEndpoint();
const agents = await buildAgents(
[remoteAgent(endpoint.url)],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
null,
);
@@ -363,7 +366,7 @@ describe("standing agent roles", () => {
await using endpoint = fakeAgUiEndpoint();
const agents = await buildAgents(
[remoteAgent(endpoint.url)],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
null,
);
@@ -389,7 +392,7 @@ describe("standing agent roles", () => {
reason: "Expense Manager has been deleted.",
},
],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
null,
);
@@ -417,7 +420,7 @@ describe("standing agent roles", () => {
seen.actors.push(actor);
return [remoteAgent("http://coworker.internal/ag-ui")];
},
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
async () => null,
);
@@ -436,7 +439,7 @@ describe("standing agent roles", () => {
async () => [
remoteAgent("http://coworker.internal/ag-ui", { roleDescription }),
],
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
async () => null,
);
const request = new Request("http://openbot.test/api/copilotkit");
@@ -581,7 +584,7 @@ describe("what a Bot is told it holds", () => {
type: "built_in",
systemPrompt: "Investigate policies.",
},
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
"openai-secret",
drive,
"BROWSER GUIDANCE HERE",
@@ -621,7 +624,7 @@ describe("where a Bot says its answer came from", () => {
type: "built_in",
systemPrompt: "Be helpful.",
},
- { provider: "openai", defaultModel: "gpt-4.1" },
+ { provider: "openai", defaultModel: "gpt-5.6-terra" },
"openai-secret",
).prompt as string;
diff --git a/server/tests/tenant-package.test.ts b/server/tests/tenant-package.test.ts
index f29499dd..40d04198 100644
--- a/server/tests/tenant-package.test.ts
+++ b/server/tests/tenant-package.test.ts
@@ -80,7 +80,7 @@ function loadedPackage(
model: {
provider: "openai",
credentialSecretRef: "openai-key",
- defaultModel: "gpt-4.1",
+ defaultModel: "gpt-5.6-terra",
},
knowledgeSources: [],
themeCss: "",
@@ -145,7 +145,7 @@ describe("tenant YAML validation", () => {
"agents: [{ id: knowledge, name: Knowledge, role_description: Answer company questions., type: built-in, system_prompt: Answer from knowledge. }]",
channels: "channels: []",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: "",
}),
@@ -160,7 +160,7 @@ describe("tenant YAML validation", () => {
"agents: [{ id: knowledge, name: Knowledge, title: Company Knowledge, type: built-in, system_prompt: Answer from knowledge. }]",
channels: "channels: []",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: "",
}),
@@ -189,7 +189,7 @@ describe("tenant YAML validation", () => {
agents: `agents: [{ id: ${reserved}, name: Knowledge, title: Company Knowledge, role_description: Answer company questions., type: built-in, system_prompt: Answer from knowledge. }]`,
channels: "channels: []",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: "",
}),
@@ -205,7 +205,7 @@ describe("tenant YAML validation", () => {
"agents: [{ id: policy-desk, name: Policy Desk, title: Policy, role_description: Answer policy questions., type: built-in, system_prompt: Answer from policy. }]",
channels: "channels: []",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: "",
});
@@ -233,7 +233,7 @@ describe("tenant YAML validation", () => {
endpoint: http://risk.internal/ag-ui`,
channels: "channels: []",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: "",
});
@@ -250,7 +250,7 @@ describe("tenant YAML validation", () => {
"agents: [{ id: knowledge, name: Knowledge, title: Company Knowledge, role_description: Answer company questions., avatar_seed: '', type: built-in, system_prompt: Answer from knowledge. }]",
channels: "channels: []",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: "",
}),
@@ -265,7 +265,7 @@ describe("tenant YAML validation", () => {
agents: "agents: []",
channels: "channels: []",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: ":root { --primary: oklch(0.32 0.09 250); }",
}),
@@ -317,7 +317,7 @@ describe("tenant YAML validation", () => {
brand: `tenant:\n id: fintech\n product_name: Ledgerline\nskin:\n stylesheet: theme.css`,
agents: `agents:\n - id: knowledge\n name: Knowledge\n title: Company Knowledge\n role_description: Answer company questions.\n type: built-in\n system_prompt: Answer from knowledge.\n - id: risk\n name: Risk\n title: Risk & Compliance\n role_description: Investigate policies and controls.\n type: remote-ag-ui\n endpoint: http://risk.internal/ag-ui`,
channels: `channels:\n - id: company\n name: Company\n description: Knowledge channel\n permitted_agents: [knowledge, risk]\n allowed_groups: [all]`,
- model: `model:\n provider: openai\n credential_secret_ref: openai-key\n default_model: gpt-4.1`,
+ model: `model:\n provider: openai\n credential_secret_ref: openai-key\n default_model: gpt-5.6-terra`,
knowledge: `sources:\n - type: google-drive\n roots: [Policies]`,
themeCss: ":root { --primary: black; }",
});
@@ -335,7 +335,7 @@ describe("tenant YAML validation", () => {
channels:
"channels: [{ id: company, name: Company, description: Test, permitted_agents: [missing], allowed_groups: [all] }]",
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: ":root { --primary: black; }",
}),
@@ -365,7 +365,7 @@ describe("tenant YAML validation", () => {
permitted_agents: [knowledge, risk-analyst]
allowed_groups: [all]`,
model:
- "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-4.1 }",
+ "model: { provider: openai, credential_secret_ref: openai-key, default_model: gpt-5.6-terra }",
knowledge: "sources: []",
themeCss: "",
});