Bland

Fetch one agent by id, including its environment pointers (which version staging/production are pinned to; dev is always live). Cross-org or missing ids fail.

Fetch an agent's check config for an environment (the judges + scenarios that gate promotions into it), or null when none is configured.

Fetch one production check run by id, including its per-eval verdicts and the linked simulation set (for viewing the generated conversations).

Fetch a specific published version (by semver, e.g. "1.2.0") including its full snapshot — used to view what an environment is pinned to.

Fetch the full markdown of a single Bland docs page (e.g. path 'api-v1/post/calls'). Use after search_bland_docs.

Fetch a single call — its transcript, recording, summary, and metadata — and render the call-log inspector. Scoped to the current organization; missing/cross-org call ids 404.

Fetch a single eval agent (an LLM judge) and its current version — criterion prompt, scoring levels, target levels, weight, and modality — and render the eval-agent card. Scoped to the current organization; missing or cross-org ids 404.

Fetch a single eval run and its per-atom verdicts, and render an eval scorecard. The run is scoped to the current organization; cross-org or missing run ids 404.

Fetch an agent's latest saved version including its full native config snapshot — what the builder loads. Fails for a fresh agent with no versions yet.

DEEP pathway understanding (read-only): the same semantics-explorer summaries the server-side pathway agent uses. Pass the raw graph ({ nodes, edges }) plus a scope: 'node' (+ node_id/node_name/node_slug) returns the FULL execution contract for one node — prompt/tool/loop/route/transfer contracts, operation order, dependency + stuck-reason inference, optionally include_neighbors; 'transition' (+ source_*/target_* selectors) returns the routing semantics for one edge — route kind, condition, decision basis, file pointers; 'dependencies' (+ node selector) or 'pathway' returns missing user config + existing resources. Use it BEFORE editing a node or debugging routing/loop behavior — validate_pathway tells you WHAT is broken, this tells you HOW the node/edge actually executes. Nothing is persisted.

Return the AUTHORITATIVE allowed shape + enums for a structured pathway surface (read-only). Call this BEFORE hand-authoring or heavily editing a structured YAML surface (tools.yaml / variables.yaml / model.yaml / unit-tests.yaml) — or a node/edge — so you get it valid first-try. This is the schema guidance the old set_* tools used to give. `surface` is one of node_tools | variables | model | unit_tests | node | edge. For surface='node_tools', optionally pass `tool_type` (webhook | custom_tool | code | track) to get just that tool-type's config variant instead of the full set_node_tools envelope. Returns { surface, description, json_schema, allowed_values? }. Nothing is persisted.

List an agent's open branches (dev workspaces). `behind` means dev's head moved past the branch's base, so merging requires a rebase first.

List an agent's check runs for an environment (newest first) — each run's status, per-eval verdicts, and the candidate version it scored.

List an environment's deployment history (default production), newest first, with each row's version name + semver and the check run that produced it. `isRollbackEligible` flags production versions that aren't the current pin — the rollback candidates.

List an agent's dev/staging/production environments with the semver each is pinned to (dev is always live/unpinned).

List an environment's variables for an agent (plain key→value pairs).

List an agent's saved version history (metadata only — no snapshots), newest first.

List the organization's agents (the /v2 agents model), most recently updated first. Soft-deleted agents are excluded.

List the organization's eval agents (LLM judges), each with its latest-run target-hit summary, and render the eval-agent list. Selecting a row opens that specific agent via get_eval_agent.

Start here for every new voice choice: outbound calls (`voice` on create_call or POST /v1/calls), personas (`call_config.voice`), inbound numbers, and pathways. Returns Bland's curated voices on its current, recommended voice generation; match the user's character brief against each voice's description and tags and use its `id`. Every result carries `service`, the same value as `model_id` on GET /v1/models, so you can tell a current-generation voice from an older one when other catalog reads are in play. If the user names a specific voice (including their own cloned voice), use it as given even when it is not listed here, and keep an existing voice during unrelated edits. Omitting `voice` on create_call uses Karen.

Merge an open branch into dev (fast-forward only): squashes the branch head into one new dev version and archives the branch. Fails if dev moved past the branch's base — rebase first.

Promote an agent to production, carrying the version's original semver. Omit versionId to pin production to whatever staging currently points at; provide versionId to take that exact version (dev timeline or open branch) live in one step, repointing staging to it as well and minting a semver if it never had one. The current saved evaluation policy is enforced unless force is set; run current required evaluations first.

Publish an agent to staging (dev→staging). Provide `snapshot` to save it as a new immutable version and pin staging; provide `versionId` to publish an existing version; provide neither to publish the dev head. A brand-new version mints the next semver for the chosen bump; a version that already has a semver re-records it. Returns the version, its semver, and updated environments.

Query call analytics data using structured parameters. Returns aggregated metrics with optional grouping, filtering, and time ranges, and renders an analytics panel. All queries are automatically scoped to the current organization.

Run a read-only shell-like query against a virtualized, in-memory filesystem rooted at `/` that contains ONLY the Bland documentation pages. This is NOT a shell on any real machine — nothing runs on the user's computer, the server host, or any network. The filesystem is a sandbox backed by documentation chunks. This is how you read documentation pages: there is no separate "get page" tool. To read a page, pass its `.mdx` path to `head` or `cat` — a page at the URL path `/some/page` lives at `/some/page.mdx`. To search the docs with exact keyword or regex matches, use `rg`. To understand the docs structure, use `tree` or `ls`. **Paths are specific to this site — never guess them.** Discover real paths with `tree / -L 2`, `ls /`, or the search tool before reading. If a path does not exist, that only means the guess was wrong; it does NOT mean the topic is undocumented — use `rg -il "keyword" /` to find where it is covered. **Workflow:** Start with the search tool for broad or conceptual queries like "how to authenticate" or "rate limiting". Use this tool when you need exact keyword/regex matching, structural exploration, or to read the full content of a specific page by path. Supported commands: rg (ripgrep), grep, find, tree, ls, cat, head, tail, stat, wc, sort, uniq, cut, sed, awk, jq, plus basic text utilities. No writes, no network, no process control. Run `--help` on any command for usage. Each call is STATELESS: the working directory always resets to `/` and no shell variables, aliases, or history carry over between calls. If you need to operate in a subdirectory, chain commands in one call with `&&` or pass absolute paths (e.g., `cd /some-directory && ls` or `ls /some-directory`). Do NOT assume that `cd` in one call affects the next call. Examples (replace the placeholder paths with real ones from `tree` or search): - `tree / -L 2` — see the top-level directory layout - `rg -il "rate limit" /` — find all files mentioning "rate limit" - `rg -C 3 "apiKey" /some-directory/` — show matches with 3 lines of context around each hit - `head -80 /some/page.mdx` — read the top 80 lines of a specific page - `head -80 /page-one.mdx /page-two.mdx /section/page-three.mdx` — read multiple pages in one call - `cat /some/page.mdx` — read a full page when you need everything Output is truncated to 30KB per call. Prefer targeted `rg -C` or `head -N` over broad `cat` on large files. To read only the relevant sections of a large file, use `rg -C 3 "pattern" /path/file.mdx`. Batch multiple file reads into a single `head` or `cat` call whenever possible. When referencing pages in your response to the user, convert filesystem paths to URL paths by removing the `.mdx` extension. For example, `/some/page.mdx` becomes `/some/page`.

Rebase a branch onto the dev head from a rebase preview: apply the three-way merge with the given per-conflict resolutions, save the result as a new branch version, and move the branch's base to that dev head. Merge afterwards.

Dry-run merging a branch onto the current dev head: three-way diff (branch base vs dev vs branch head) listing the dev-side changes that merge in cleanly and the fields that conflict. Read-only.

Rename an agent (the /v2 agents model).

Roll an agent's production back to a previously-deployed version: repoint production and re-record that version's original semver (never mints). Fails if the version was never deployed to production or is already the current pin.

Save an agent's native config as an immutable version snapshot (the builder's Save). Versions carry no number — a semver is minted when the version is published; naming a version pins it forever. The snapshot must be the full {behavior, settings, contact} agent config.

Search across the Bland knowledge base to find relevant information, code examples, API references, and guides. Use this tool when you need to answer questions about Bland, find specific documentation, understand how features work, or locate implementation details. The search returns contextual content with titles and direct links to the documentation pages. If you need the full content of a specific page, use the query_docs_filesystem tool to `head` or `cat` the page path (append `.mdx` to the path returned from search — e.g. a result at `/some/page` is read with `head -200 /some/page.mdx`).

Search the Bland API documentation (docs.bland.ai). Returns matching doc pages (title, url, description). Use get_bland_doc to read a page in full, then use a tool available in this connection that supports the requested operation.

Create or fully replace an agent's production check config: which evals score the candidate (with their passing levels + required flag — a required judge blocks the promotion when it fails), which scenarios drive the simulated conversations, and how many conversations per scenario. Validates that scenarios belong to the agent and every eval version is runnable.

Create or update one environment variable for an agent. Set isSecret to store the value in the secret store and keep only a {{SECRET.NAME}} reference on the variable.

Start a standalone check run: the environment's configured judges score simulated conversations against a candidate version (default: what a promotion into this env would deploy). Poll get_agent_check_run until terminal — PASSED means every required judge passed. Runs never move environment pointers; gate your own publish/promote on the outcome.

End a call that is in progress, or cancel one that is still queued. This hangs up on the person being called, so confirm with the user first. Scoped to the current organization.

Report a problem with this documentation site so the docs team can fix it. Use when a documentation page is incorrect, outdated, confusing, incomplete, or has a broken example. This is for feedback about the documentation content itself — not for product support requests or feedback about this tool or assistant.

AUTHORITATIVE Bland pathway compiler (read-only). Pass the raw graph you are about to commit ({ nodes, edges }, optional requestData) and it runs the full Layer-2 compile: start/end/reachability/dead-end structural checks, routing/loop/dialogue/tool-input runtime contracts, and per-node semantics. Returns { valid, errors, warnings, stats, runtime_contract_findings, semantics_summary }. Nothing is persisted — validate BEFORE committing. CHANGE-AWARE: additionally pass `baseline` (the PRE-EDIT graph, e.g. from .norm/baseline.json) to diff your edit against it — the response then also carries `introduced_errors`, `introduced_warnings` and a `validation_delta_summary` (current vs introduced vs pre-existing counts), and the runtime_contract_findings your edit caused are flagged relevant_to_changes=true ([NEW FROM YOUR CHANGES]) so you can tell what YOU broke from what was already broken. Omit `baseline` for plain whole-graph validation.

Wait for a call to finish, then return its call-log — use this instead of polling get_call_log in a loop. Holds the request open for up to `timeoutSeconds` (default 30, max 45) and returns the moment the call reaches a terminal state. The call is finished when `status` is one of: completed, failed, error, no-answer, busy, canceled, cancelled — or when `errorMessage` is set. Timing out is not an error: you get the call-log as it stands, and a call-log showing none of those is still running — call again to keep waiting. Scoped to the current organization.