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.exampleindex 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 adiff --git a/agent-langgraph/src/index.ts b/agent-langgraph/src/index.tsindex 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.ymlindex 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.mdindex 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```shOPENAI_BASE_URL=https://gateway.internal/v1OPENAI_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: openaicredential_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: openaicredential_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.yamlindex 5bdce6a2..2c0d518a 100644--- a/examples/fintech/model.yaml+++ b/examples/fintech/model.yaml@@ -1,4 +1,4 @@model:provider: openaicredential_secret_ref: openai-api-key- default_model: gpt-4.1+ default_model: gpt-5.6-terradiff --git a/examples/langgraph-bot/src/index.ts b/examples/langgraph-bot/src/index.tsindex 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.tsindex 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.tsindex 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.tsindex 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: "",});